Revista Innovación y Software

Vol. 6, No. 2, Mes Septiembre - Febrero, 2025

ISSN: 2708-0935

Pág. 134-158

https://revistas.ulasalle.edu.pe/innosoft

Esta obra está bajo una Licencia

Creative Commons Atribución

4.0 Internacional.

Tipo de artículo: Artículos originales

Temática: Redes y seguridad informática

Recibido: 21/02/2025 | Aceptado: 21/03/2025 | Publicado: 30/09/2025

Identiﬁcadores persistentes:

DOI: 10.48168/innosoft.s24.a317

ARK: ark:/42411/s24.a317

PURL: 42411/s24.a317

A Case Study of the Salsa20 Encryption Algorithm Using

Random Noisy Injection Enhanced by Artiﬁcial Intelligence

A Case Study of the Salsa20 Encryption Algorithm Using

Random Noisy Injection Enhanced by Artiﬁcial Intelligence

Edgar Rangel Lugo1[0000-0002-9611- 8323]*, Kevin Uriel Rangel Ríos2[0009- 0001- 7029- 7183], Carlos Alberto

Bernal Beltrán3[0009-0009- 3244- 8268], Leonel González Vidales4[0000- 0002- 2623- 7626], César Del Ángel

Rodríguez Torres5[0000-0001-9198-4372], Lucero De Jesús Ascencio Antúnez6[0009- 0008- 0923-4707], Rosa

Isabel Reynoso Andrés7[0009-0001-8896-8328]

1Tecnológico Nacional de México. I.T. de Ciudad Altamirano. Av. Pungarabato Ote. S/N, Col. Morelos,

Ciudad Altamirano, Guerrero, México. erangel_lugo@hotmail.com

2Tecnológico Nacional de México. I.T. de Ciudad Altamirano. Av. Pungarabato Ote. S/N, Col. Morelos,

Ciudad Altamirano, Guerrero, México. kgvppro@gmail.com

3Tecnológico Nacional de México. I.T. de Ciudad Altamirano. Av. Pungarabato Ote. S/N, Col. Morelos,

Ciudad Altamirano, Guerrero, México. carlosalberto.bb@cdaltamirano.tecnm.mx

4Tecnológico Nacional de México. I.T. de Ciudad Altamirano. Av. Pungarabato Ote. S/N, Col. Morelos,

Ciudad Altamirano, Guerrero, México. leonel.gv@cdaltamirano.tecnm.mx

5Tecnológico Nacional de México. I.T. de Ciudad Altamirano. Av. Pungarabato Ote. S/N, Col. Morelos,

Ciudad Altamirano, Guerrero, México. cesardelangel.rt@cdaltamirano.tecnm.mx

6Tecnológico Nacional de México. I.T. de Ciudad Altamirano. Av. Pungarabato Ote. S/N, Col. Morelos,

Ciudad Altamirano, Guerrero, México. lucerodejesus.aa@cdaltamirano.tecnm.mx

7Tecnológico Nacional de México. I.T. de Ciudad Altamirano. Av. Pungarabato Ote. S/N, Col. Morelos,

Ciudad Altamirano, Guerrero, México. rosaisabel.ra@cdaltamirano.tecnm.mx

∗Autor para correspondencia: erangel_lugo@hotmail.com

Resumen

La pérdida o robo digital de datos, puede generar severas consecuencias ﬁnancieras para las organizaciones.

Dicha situación puede ser manejada utilizando métodos de cifrado dinámicos, los cuales, son capaces de generar

diferentes textos cifrados partiendo de una misma entrada de texto plano. La presente investigación, involucra

el uso de inteligencia artiﬁcial (IA), aplicado a la ciberseguridad, a través del uso de estrategias: .aleatorias

ruidosas"(random noisy), siendo estudiados varios aspectos relacionados en dicho campo. La aplicación de estos

esquemas: random noisy, sobre el texto cifrado, es recomendado como medida para mejorar la seguridad de los

datos digitales. Estudios recientes reﬁeren que ello puede ser efectivo para contrarrestar posibles ciberataques.

Por lo tanto, el objetivo de la presente investigación fue enfocado sobre la aplicación de una nueva propuesta

basada en estrategia: random noisy, siendo aplicada a textos cifrados obtenidos por el algoritmo Salsa20, debido

a que, ha sido poco estudiado en la literatura. Lo anterior, sustentado en que las estrategias .aleatorias ruido-

sas", han revelado resultados muy prometedores, en previas investigaciones. El presente trabajo, se considera

importante, debido a que, los resultados experimentales han recomendado la efectividad del uso de IA, aplicado

Facultad de Ingeniería

Universidad La Salle, Arequipa, Perú

facin.innosoft@ulasalle.edu.pe

134

Revista Innovación y Software

Vol. 6, No. 2, Mes Septiembre - Febrero, 2025

ISSN: 2708-0935

Pág. 134-158

https://revistas.ulasalle.edu.pe/innosoft

en la inyección de ruido a textos cifrados, mostrando incremento en el esquema de seguridad de los datos digita-

les encriptados. Por lo tanto, es presentada aquí una nueva metodología mejorada, para las organizaciones que

utilizan cifrado de datos con el algoritmo Salsa20, así como, la comparación de resultados con cuatro random

noisy.estrategias, basadas sobre el cifrado con los algoritmos: DES, 3DES, AES-256, y Blowﬁsh. En conclusión,

la nueva alternativa, aquí presentada, es recomendada porque mostró ser de difícil desencriptado, incluso para

las herramientas estándar de los ciber-criminales.

Palabras claves: Aplicaciones de IA, criptografía, cifrado de datos dinámico, cifrado con inyección de ruido.

Abstract

The loss of digital data can have severe ﬁnancial consequences for organisations. This situation can handle with

a dynamic encryption approach, which it can generate multiple ciphertexts from a single plaintext. This paper

delves into the intersection of artiﬁcial intelligence, random noisy schemes, and cybersecurity, analyzing several

critical aspects of this emerging ﬁeld. The application of random noisy scheme to ciphertext, it has been suggested

as a means of enhancing cybersecurity, with recent studies indicating that this approach can be eﬀective against

cybercriminals. The objective of this research was focused on applying random noisy strategy on ciphertext of

the Salsa20 encryption algorithm, as this area remains unexplored. Given the promising results of random noisy

strategies, organisations that they employ Salsa20 may beneﬁt from this approach. This research is important

because recent experiments have revealed the eﬀectiveness of using artiﬁcial intelligence for noisy injection on

ciphertext. Therefore, is here presented a new opportunity for organisations regarding the employment of the

novel random noisy Salsa20 encryption alternative because the experimental results have shown an increasing

of the dynamic performance on ciphertext. This work introduces the random noisy Salsa20 strategy as a novel

dynamic encryption alternative, and comparison of the performance to four random noisy schemes based on

DES, 3DES, AES-256, and Blowﬁsh algorithms. In conclusion, the novel alternative that it is here recommended,

it can be diﬃcult for the cybercriminals to decrypt.

Keywords: Applications of AI, cryptography, dynamic encryption methods, noisy injection strategies.

Introduction

A cybersecurity strategy [1–3], it is deemed insuﬃcient if even one of its methods is susceptible to cyberattacks

[4–8], which can lead to the theft of sensitive digital data [8,9]. In practical applications, this can result in

substantial ﬁnancial losses for organisations [4,5,8,10,11]. These cases [12–17] typically involve cyber threats

such as phishing, fraudulent calls, and ﬁnancial scams, impacting social networking platforms, bank systems,

large markets, energy networks, and e-commerce in organisations [4,5]. Several approaches have been proposed

to tackle the issue of digital data theft. These can be broadly categorised into three main strategies: Firstly,

regularly updating cybersecurity strategies [4]. Secondly, utilising dynamic encryption methods [5]. Thirdly,

employing noisy injection strategies on ciphertext [5–7,9,10], which they have shown promise in certain practical

domains. This study focuses on encryption methods [8], that they utilise noisy injection strategies [5–7]. These

methods involve computing statistics or mathematical equations to transform plaintext (Si) into ciphertext

(Ci) [8], which can only be deciphered by authorised individuals or systems [10,11]. The Si represents the

original input data, while the Ci is the encrypted output. An encryption method is dynamic if it produces

unique results with the same plaintext data input each time, whereas static encryption methods produce the

Facultad de Ingeniería

Universidad La Salle, Arequipa, Perú

facin.innosoft@ulasalle.edu.pe

135

Revista Innovación y Software

Vol. 6, No. 2, Mes Septiembre - Febrero, 2025

ISSN: 2708-0935

Pág. 134-158

https://revistas.ulasalle.edu.pe/innosoft

same output [8]. Symmetric encryption algorithms, they use a single secret key, while asymmetric encryption

algorithms use a pair of keys, one private and one public [8]. Data encryption refers to the conversion of plaintext

into ciphertext, whereas data decryption is the reverse process [8–10,14,18]. Asymmetric algorithms like RSA

(Rivest-Shamir-Adleman) [3,5,6,8,13,14,19–27], ECC [5,6,8,19–23,28–31], and ElGamal [13,19–21], they

use asymmetric key cryptography, which involves a pair of keys - private and public. Dynamic encryption has

shown promising results in recent research [5,8,11], when applied to asymmetric algorithms. Since asymmetric

algorithms are not within the scope of this work, they are not examined here and could be considered for

future works. There are also various symmetric key cryptography algorithms, such as DES (Data Encryption

Standard) [3,13,14,22,23,27,32,33], 3DES or TripleDES (Triple Data Encryption Standard) [14,21–23,34],

Blowﬁsh [22,23,35–38], Salsa20 [38,39], and AES (Advanced Encryption Standard) [13,21–23,26,27,40,41],

to mention a few. In this research, the AES-256 version [38,40,41], it has been employed. In reference [22], it

is noted that AES is a symmetric block cipher that can operate with diﬀerent block sizes and supports key

lengths of 128, 192, and 256 bits. DES processes 64 bits of plaintext to produce 64 bits of ciphertext, using a

series of substitution and permutation rounds, and decryption is achieved by reversing the encryption process.

Due to the limited key size of 64 bits, DES is considered insecure for modern standards. To improve security,

3DES was developed as a more robust alternative. Another symmetric algorithm, Blowﬁsh, employs a variable-

length block cipher with key lengths spanning from 32 to 448 bits [22,23]. Although Blowﬁsh is versatile, it

is often used with a block size of 64 bits. Regarding the symmetric algorithm Salsa20 [39], it is ChaCha20’s

predecessor. Both algorithms are not block-cipher because they work with keystream generation. According

to [39], Salsa20 is addition/rotation/XOR (ARX)–based cryptographic primitive. Their keystream generation

algorithm comprises three simple operations: Addition modulo 232 (⊞), constant distance left bit rotation

(n), and bitwise XOR operation. These operations are swift in any circuit, and hence the cipher can achieve

signiﬁcant speed with a high security margin. The Salsa20 cipher takes a 256-bit key (k), a 128-bit constant

(c), and a 128-bit initial vector (v), or IV, and generates a 512-bit keystream. The key is divided into eight

32-bit words (k0, k1, . . . , k7). Similarly, the constant and the IV are also broken into 32-bit words (c0, c1, c2, c3)

and (v0, v1, t0, t1). Due to the use of the 256-bit key, the cipher is also called 256-bit Salsa20. In the original

Salsa20 cipher, the number of rounds is 20. After the ﬁnal round, the initial state Xis added modulo (232) to

the updated state X20 word-by-word, and the keystream (Z) of 512 bits is achieved, that is, Z=X⊞X20. This

keystream is then bitwise XORed to plaintext to get the ciphertext. Besides, the Salsa20 scheme, it is commonly

generated a 512-bit keystream. This proposal is appealing for encryption due to its high speed and security. Due

to their eﬃcient algorithm and fast performance, it has been attracted by cryptographic analysis since their

release [39]. Some research [5,8], they have revealed that symmetric encryption algorithms, they have reported

static ciphertext results. Despite that it does not mean, in all cases that they are vulnerable schemes [5].

However, there exists some standard encryption alternatives that they have been threatened by quantum

computing [8,22,23,42]. In contrast to some views, other researchers [43–45], they emphasize that the AES and

Facultad de Ingeniería

Universidad La Salle, Arequipa, Perú

facin.innosoft@ulasalle.edu.pe

136

Revista Innovación y Software

Vol. 6, No. 2, Mes Septiembre - Febrero, 2025

ISSN: 2708-0935

Pág. 134-158

https://revistas.ulasalle.edu.pe/innosoft

RSA algorithms are at risk due to advancements in quantum computing [8,22,42]. This paper investigates several

facets of this issue and, without using quantum computing, proposes measures that can be applied to symmetric

algorithm variants to improve their security [5]. For example, as pointed out by some authors [4–11], the dynamic

encryption methodologies, they oﬀer a good alternative to increasing noisy and redundant information in

ciphertext outputs [5,9,10]. Moreover, the employment of noisy injection strategies [5–7], they are recommended

as good encryption measure. The noisy injection means that a plaintext or ciphertext contains some additional

characters of the ASCII or UTF-8 encoding, which they are not part of the original input message [8]. A key

hypothesis of this research is that these strategies can create confusion for cybercriminals, as indicated in [5,8].

These strategies [5–7], frequently utilize artiﬁcial intelligence (AI) [5–8,10,11,46,47], as AI-based cryptography

[1–4,8,10,11,16,17,24,48–50], it is a well-established ﬁeld. According to [10,46,47], the purpose of AI is to

enable machines to think [8]. In this context, various approaches like heuristic methods [5,8,10], they have

been proposed, which involve predeﬁned rules to solve problems. These can be applied to implement structured

models such as decision trees [51–53] and graphs [53–55], among others. Moreover, there exists alternatives with

AI based on random methods [10,56–59], it consists in the selection of random numbers without replacement

or with replacement (i.e., it does not repeat or it repeats values, respectively) [8]. These methodologies can be

employed with intelligent models such as genetic algorithms (GAs) [1,2,9,10,16,17,56,59–62], Monte Carlo

(MC1) algorithm [59], artiﬁcial neural networks [57,63,64], to mention few. Several authors have explored

AI-based cryptography alternatives [1–4,8,10,11,14,16,17,24,27,48–50,60–62]. For a comprehensive review of

AI applications in cryptography, see [50]. In [48], it is analyzed trends and challenges, while in [3], it focuses

on asymmetric and symmetric cryptographic methods. In [58], a study on pairing functions for AI-driven

cryptography was conducted. Subsequently, the same author [16], he published a novel investigation on GAs

in cryptography, speciﬁcally contributing to the ﬁeld of e-commerce. In addition, to existing research, in [24]

reports on an improved cryptanalysis of RSA using side-channel attacks. Meanwhile, reference [49], it employs

a new approach to collision attacks through diﬀerentiated path construction. In line with established research

practices, genetic algorithms for cryptography have been explored in [1,62]. Similarly, in [60], it introduces

modern optimization algorithms designed for cryptanalysis. On the ﬂip side, reference [61], it reveals that genetic

algorithms can successfully break certain simple cryptographic ciphers. In a similar vein, it is examined in [27],

the application of genetic operators to symmetric cryptography using GAs. Moreover, in [49], it is introduced

a quantum cryptanalysis technique for lattice-based protocols. Concerning the noisy injection strategies [4–11],

there exists several alternatives that they can be grouped in several categories. Initially, pseudo-hexadecimal

format is utilized. The “Noised” random pseudo-hexadecimal GAs methodology, in particular, it has been

discussed in references [9,10]. A dynamic encryption scheme based on genetic algorithms was proposed. However,

because of the disadvantages associated with pseudo-hexadecimal GAs, a successor methodology, “Noised”

random pseudo-hexadecimal (without GAs), it was presented in [10]. Later, in [8], it presented four dynamic

alternatives based on the pseudo-hexadecimal schema, known as «noisy random pseudo-hexadecimal» strategies.

Facultad de Ingeniería

Universidad La Salle, Arequipa, Perú

facin.innosoft@ulasalle.edu.pe

137

Revista Innovación y Software

Vol. 6, No. 2, Mes Septiembre - Febrero, 2025

ISSN: 2708-0935

Pág. 134-158

https://revistas.ulasalle.edu.pe/innosoft

The goal of these strategies is to confuse cybercriminals by injecting noise into ASCII characters when utilizing

a novel pseudo-hexadecimal format. These schemes are limited to plaintext applications. The second approach

involves combining AI-based noisy injection with the 1-NN rule, as discussed by other research [46,47,65–67]. In

this regard, the Ñoised’random 1-NN with hexadecimal encoding based on AI was introduced in [11]. Similarly,

the combination of “Noised” random pseudo-hexadecimal format with 1-NN was explored in [68]. These two

approaches have shown that implementing double noisy injection over ciphertext outputs can improve data

safety, though it requires sacriﬁcing some disk storage space. These approaches were also used with plaintext.

The pseudo-hexadecimal scheme, it was not examined in this paper, but it might be addressed in future

works. Thirdly, the random Caesar II mod 120 [4–7,11], it is utilized for noisy injection, applied to both

plaintext [4,5,11], and ciphertext [5–7], which it was obtained through standard encryption algorithms. This

method is referred to as the random noisy strategy [5–7]. Reference [5], it highlights that the random Caesar

proposal, based on the traditional Caesar algorithm [4,9,13,14], stands out due to its dynamic encryption

with AI, utilizing heuristic methods. While the traditional Caesar cipher relies on a ﬁxed shifting value K,

as expressed in Ci =Si +K(m´od 26) [13,14], the random Caesar utilizes varying shifting values (Ki) for

each character Si, chosen randomly with replacement. The operation: Ci =Si +Ki (m´od N), it allows

random Caesar to adapt to diﬀerent Nvalues. In the case of the traditional Caesar cipher with (m´od 26),

the alphabet is restricted to 26 characters for encryption and decryption purposes. While the (m´od N)in

random Caesar, it can be dynamic, recent studies [4,6,9,11], they have focused on three speciﬁc modes:

random Caesar I with (m´od 9) and (m´od 255) [6], random Caesar II with (m´od 95) [4,9,11], and random

Caesar II with (m´od 120) [4–7]. The selection of alphabets in these schemes is deﬁned by random conditions,

and a heuristic method is applied to ﬁnd the most eﬀective Ki vector. Here, the (m´od N)value indicates

the size of the encryption alphabet and the maximum value that can be present in the Ki vector. With

N= 95, ASCII characters from space (32) to ’∼’(126) are used. N= 120 covers a range from 30 to 150 in

the ASCII table. For diﬀerent Nvalues, Ki falls between 0 and N. The random Caesar scheme’s procedure

includes a second phase intended to confuse potential attackers [4]. This phase calculates the ﬁnal package

as F inalP ackage =Ci &Ki &OrdChr(Ci)[5]. The &operator indicates concatenation, and the OrdChr

procedure either appends the character Ci when N= 120 or converts it to an ordinal value depending on

the mode. This method is limited to plaintext encryption as a dynamic strategy [4], while random Caesar II

mod 120 has been employed in some studies [5–7], as a means of introducing randomness. Reference [5], it

presents eight alternatives that integrate random noise with standard encryption algorithms: random noisy

DES, random noisy 3DES, random noisy RC4, random noisy Blowﬁsh, random noisy WEP, random noisy

AES, random noisy RSA-2048, and random noisy ECIES SECP-256-R1. The approach integrates standard

encryption techniques with random Caesar II mod 120 for noisy injection. Similarly, the concept of random

noisy GOST was explored in [6], and the random noisy Camellia was highlighted in [7]. Some variants of random

noisy strategies [4], they focus on noisy injection and they were designed to camouﬂage ciphertext [5]. Notable

Facultad de Ingeniería

Universidad La Salle, Arequipa, Perú

facin.innosoft@ulasalle.edu.pe

138

Revista Innovación y Software

Vol. 6, No. 2, Mes Septiembre - Febrero, 2025

ISSN: 2708-0935

Pág. 134-158

https://revistas.ulasalle.edu.pe/innosoft

examples are reduced random Caesar [4,5], and reduced random mutation [4]. According to [4], the primary

purpose of these schemes is to camouﬂage and reduce the size of at least 1/3of the ciphertext. The reduced

random Caesar strategy has been applied to both plaintext [4], and ciphertext [5], whereas the reduced random

mutation [4], it has been used exclusively with plaintext. These reduced random and mutation strategies, they

were not here experimented because it can be considered in future works. We assessed four random noisy

strategies that they utilize standard symmetric encryption algorithms, including random noisy DES, random

noisy 3DES, random noisy AES-256, and random noisy Blowﬁsh. Due to that novel alternative, named here as

random noisy Salsa20, it was developed and it is introduced in this study. The application of noisy injection

with random noisy strategies [4–7], to plaintext [4], or ciphertext [5–7], they are recommended due to its strong

correlation with dynamic encryption performance. There exists a lack of research on the vulnerabilities of

these schemes when subjected to quantum computing [5,8]. Given that random noisy strategies have shown

promise, this research continues the work of [5–7], by examining the Salsa20 encryption algorithm’s potential

when applied to ciphertext, a gap in existing research that could beneﬁt organizations employing Salsa20. This

circumstance also allows organizations to consider employing noisy injection techniques derived from the Salsa20

scheme. This work explores the use of noisy random encryption in cybersecurity, speciﬁcally the technique

of applying noisy injection to ciphertext from standard algorithms to potentially deceive cybercriminals [8].

The scope of this study was limited to two classes of situations. The initial step involved comparing ﬁve

standard encryption algorithms (DES, 3DES, AES-256, Blowﬁsh, and Salsa20) as static encryption schemes,

with outcomes benchmarked against related research [5–7]. Furthermore, the random noisy Salsa20 scheme, it

was developed to serve as a comparative benchmark for other random noisy strategies (random noisy DES,

random noisy 3DES, random noisy AES-256, and random noisy Blowﬁsh) in the context of noisy injection

on ciphertext. This method uses random Caesar II mod 120 [4], being applied on ciphertext by standard

encryption algorithms. The focus is on dynamic encryption as a means to achieve randomness in performance.

It also makes an empirical contribution to cryptography using AI, as the literature on random noisy strategies

is sparse. Despite that recent research [4,5,9–11], they have reported that random Caesar II [4], it might

obtain random values for the ciphertext that they can be selected out of the range of the ASCII table [5]. In

practical domains that they were evaluated with random noisy strategies [5], these problems have not been

observed. The random noisy encryption strategies were previously assessed with ﬁve-fold cross-validation [5–7].

This work expresses their performance in terms of global average or accuracy [46,47,65,66]. We present the

ﬁndings of an extensive research project investigating digital data theft, including experimental results. This

study analyzed cases where inadequate static encryption methods were used, causing security concerns [5].

Initially, the experiments were focused on replacing of the static encryption schema for recommending the

random noisy strategies as dynamic encryption alternative [4]. Several ciphertext exemplars resulting from

random noisy encryption schemes are shown here. These approaches were evaluated over ﬁve samples when a

novel modiﬁcation of cross-validation [4,5,10], it has been employed. Based on observations, noisy injection over

Facultad de Ingeniería

Universidad La Salle, Arequipa, Perú

facin.innosoft@ulasalle.edu.pe

139

Revista Innovación y Software

Vol. 6, No. 2, Mes Septiembre - Febrero, 2025

ISSN: 2708-0935

Pág. 134-158

https://revistas.ulasalle.edu.pe/innosoft

ciphertext output is recommended due to its excellent performance as a dynamic encryption indicator. This

paper concludes by introducing the random noisy Salsa20 strategy, a novel dynamic data encryption approach.

In addition, the comparison of results with other four random noisy schemes based on DES, 3DES, AES-256,

and Blowﬁsh, algorithms, they are also introduced. Finally, this work is justiﬁed because the novel random

noisy Salsa20 is rarely studied, thereby it has not revealed vulnerabilities in literature regarding to cyberattacks

or quantum computing strategies.

Computational methodology

The use of static encryption algorithms as a replacement for existing cybersecurity strategies does not ensure

data protection for organizations. A dynamic encryption approach is suggested instead [4]. We evaluate several

dynamic encryption approaches that utilize random noisy strategies [5], to address the issue of digital data

theft. This research, it is considered experimental and exploratory. Due to that the novel alternative that it

is based on random noisy Salsa20 scheme, it is here ﬁrst introduced. This work required the use of hardware,

software, and datasets. The experiments were conducted on a personal computer with a 2 GHz CPU, 4 GB of

RAM, and 32 GB of free disk space. The software implementation of the encryption methods, including DES,

3DES, AES-256, Blowﬁsh, and Salsa20, as well as the novel variants based on random noisy strategies [5], these

experiments were carried out using Microsoft Windows 10 [69], and Python language [70]. To compare our

results with those in [5–7], we repeated some experiments on a mobile computing device with the same hardware

features as the personal computer mentioned above, but with Android 9 operating system [71], as well as the

PyDroid3 app [72], for software development. Our experiments showed no signiﬁcant diﬀerences. Our datasets

were composed of training samples (TS) [6,7,46,47,65], which included 1000 randomly created exemplars.

Each row in the dataset features a pattern with ﬁve columns or features. The data contains encryption and

decryption details, including ciphertext (Ci) represented as a pair of exemplars (T est1, T est2), as well as

metrics like encryption time (T C), decryption time (T D), error rate, and class label. The pattern format is

T P = [(T est1, T est2), T C, T D, Error, Label], which it allows for direct comparison with other studies [7]. In

Table 1, we display two ciphertext exemplars for the label "#Welcome#?". The label feature corresponds to the

plaintext (Si), which it is simulating a password and it includes noisy characters ’#ánd ’?’(with ordinal values

9619 and 65533, respectively). The encryption and decryption times, they were calculated in milliseconds, and

the T C,T D, and error features were represented as double precision values. The encryption strategy converts

the plaintext sequence into a ciphertext result, represented as a tuple (T est1, T est2), while T C is computed,

allowing us to observe the dynamic nature of the encryption results. During decryption of the ciphertext

sequence, T D was calculated. Both sequences are then stored in T S, including their respective T D, T C, and

error rates, using a predeﬁned structured pattern. The error rate was calculated based on the number of

characters with errors. If the ciphertext obtained through the encryption strategy does not match the plaintext

Facultad de Ingeniería

Universidad La Salle, Arequipa, Perú

facin.innosoft@ulasalle.edu.pe

140

Revista Innovación y Software

Vol. 6, No. 2, Mes Septiembre - Febrero, 2025

ISSN: 2708-0935

Pág. 134-158

https://revistas.ulasalle.edu.pe/innosoft

(Si), the error rate is determined by the proportion of Si that contains errors. In these terms, if for a ciphertext

of eight characters, its correspondent plaintext was also of eight characters, thereby an error value of 0.5, it

means that four characters from T est1or T est2, they have not be decrypted properly, and so on. The ciphertext

and plaintext are represented as sequences of characters using ASCII or UTF-8 encoding, and each sequence is

limited to a maximum of 255 characters. The 3DES and Blowﬁsh algorithms were exceptions, as they couldn’t

support sequences of 255 characters. Consequently, Blowﬁsh was experimented with up to 13 characters, and

3DES with up to 22 characters. The encryption times being evaluated mean that the comparison of algorithms

is not entirely on equal terms. To address this, the study experimented with ciphertexts ﬁlled with random

hexadecimal values to provide a more equitable assessment. This information was utilized to transform it into

a new format based on cross-validation modiﬁcations [4–7]. The updated T S format was then applied to each

encryption strategy individually. To enhance result interpretation, thereby, the Table 1 presents the arithmetic

mean and standard deviation for a plaintext with the value "#Welcome#?", which it has been processed using

most encryption algorithms under identical conditions. Finally, separate T S were created for each encryption

algorithm tested, including those with random noisy strategies.

First adopted strategy, it involves utilizing standard symmetric encryption algorithms, speciﬁcally DES, 3DES,

AES-256, Blowﬁsh, and Salsa20. They were experimented as static encryption schemes being applied over

plaintext for results comparison with other research [5–7]. We carried out these experiments on ﬁve training

samples containing the mentioned information, assessing each encryption algorithm individually through a mo-

diﬁed cross-validation approach [4,5]. These standard encryption algorithms were implemented in Python [70],

utilizing libraries such as cryptography [73], pycryptodome [74,75], and pycrypto/pycryptor, which they were

installed as part of the process. Therefore, it needs to be imported into the source code as follows: from cry-

ptography.hazmat.primitives import padding ; from cryptography.hazmat.primitives.ciphers import Cipher, algo-

rithms, modes ; from cryptography.hazmat.backends import default_backend, from Crypto.Cipher import DES,

and from Crypto.Cipher import Salsa20. Thereby, it is assumed that Si ="#Welcome#?. Using the DES algo-

rithm, the ciphertext can be computed according to the following operation: Ci = ((A(Key.encode(), M ode)).encrypt(plaintext)).hex(); print(Ci).

In the case of ciphertext produced by the 3DES and Blowﬁsh algorithms, the computation source-code is: Ci =

(Ri.update(plaintext) + Ri.finalize()).hex(); print(Ci). Similarly, the ciphertext for the AES-256 encryption

alternative can be computed as follows: Ci = (IV.encode()+(Ri.update(plaintext)+Ri.finalize())).hex(); print(Ci).

The ciphertext generated using the Salsa20 algorithm is presented below: Ci = ((A(Key.encode(), Nonce.encode())).encrypt(paintext)).hex(); print(Ci).

Where, the Acomponent refers to the used algorithm, while that the encoded Key parameter corresponds

to the secret key. The IV value, it is the initialization vector. The Mode argument, it is a valid opera-

tion mode of the algorithm. The Nonce argument, it is a valid nonce value of the Salsa20 algorithm. The

plaintext argument is the encoded Si. The encrypt() procedure returns ciphertext object, which is a class

component. The Nvalue indicates the maximum number of bytes in a character sequence. The Ri com-

Facultad de Ingeniería

Universidad La Salle, Arequipa, Perú

facin.innosoft@ulasalle.edu.pe

141

Revista Innovación y Software

Vol. 6, No. 2, Mes Septiembre - Febrero, 2025

ISSN: 2708-0935

Pág. 134-158

https://revistas.ulasalle.edu.pe/innosoft

ponent represents a partial ciphertext object that is either unpadded or incomplete. The Qi component

denotes the padding used. The updated() and ﬁnalize() procedures allow for the completion of the encry-

ption operation. In the end, the hex() function converts byte values to their corresponding hexadecimal re-

presentations. In these terms, some valid values for obtaining a ciphertext with DES algorithm, it can be

computed as follows: Key = "00000001"; A = DES.new ; N = 8 ; Mode = DES.MODE_ECB ; plain-

text = Si.encode() + (b"\x00"* (N-len(Si.encode()) % N )). The valid parameters for computing ciphertext

with the Blowﬁsh algorithm can be obtained through: Key = "00000001"; A = algorithms.Blowﬁsh ; N =

16; Mode = modes.ECB() ; Ri = Cipher(A(Key.encode()), Mode , default_backend() ).encryptor() ; plain-

text = ( Si + + str(.join( [ for k in range(0,int(N-len(Si.encode()))) ] )) ).encode(). To produce cipher-

text with 3DES, the valid values can be obtained through: Key = "000000000000000000000001"; A = algo-

rithms.TripleDES ; N = 24 ; Mode = modes.ECB() ; Ri = Cipher(A(Key.encode()), Mode , default_backend()

).encryptor() ; plaintext = ( Si + + str(.join( [ for k in range(0,int(N-len(Si.encode()))) ] )) ).encode(). Si-

milarly, the AES-256 encryption version requires the following valid parameters for ciphertext generation: Key

= "00000000000000000000000000000001"; A = algorithms.AES ; N = 256 ; IV= "0000000000000001"; Mode

= modes.CBC(IV.encode()) ; Ri = Cipher(A(Key.encode()), Mode , default_backend() ).encryptor() ; Qi =

padding.PKCS7(N).padder() ; plaintext = Qi.update( Si.encode() ) + Qi.ﬁnalize(). For the Salsa20 algorithm,

the valid parameters for ciphertext generation are: Key = "00000000000000000000000000000001"; N = 32 ;

Nonce = "00000001"; paintext = Si.encode() ; A = Salsa20.new ; Mode = None. Each of these statements

needs to be included in the source code before invoking Ci, as applicable.

Second adopted strategy, it consists in application of random noisy alternatives [5], for dynamic data encryption.

In [4], it is recommended because that these schemes can provide a good alternative for increasing the noise

in ciphertext outputs. The strategies outlined in [5], they were applied to ciphertexts produced by standard

encryption algorithms, speciﬁcally examining four cases: random noisy DES, random noisy 3DES, random noisy

Blowﬁsh, and random noisy AES-256. Due to that this study also introduces random noisy Salsa20 as a novel

approach. We implemented the ﬁve random noisy strategies in Python [70], and they were tested using a noisy

injection application that combines random Caesar II mod 120 with the ciphertext from each standard encry-

ption algorithm. This allowed for a direct comparison with other studies [5–7]. Each encryption algorithm was

evaluated separately on the ﬁve T S using an iterative process with ﬁve repetitions of cross-validation [4,5]. We

applied modiﬁed cross-validation to calculate the global average and standard deviation for each encryption

strategy. The novel proposals, as described in [5], involve noisy injection into ciphertext, and the procedu-

re for computing random noisy strategies is detailed in [5–7]. We computed it as follows: RandomNoisyi =

Char(Ord(StandardEncryptioni)+Ord(Ki)) & Char(Ki) & Char(StandardEncryptioni), with (m´od 120).

By replacing the plaintext with ciphertext, the operation was streamlined, as shown in [4]. The adjustment

to (m´od 120) is made because F inalP ackage,orRandomNoisyi, only stores character types. As discussed

Facultad de Ingeniería

Universidad La Salle, Arequipa, Perú

facin.innosoft@ulasalle.edu.pe

142

Revista Innovación y Software

Vol. 6, No. 2, Mes Septiembre - Febrero, 2025

ISSN: 2708-0935

Pág. 134-158

https://revistas.ulasalle.edu.pe/innosoft

in [5], several versions of random noisy schemes have been explored. In this research, only four strategies for

obtaining the StandardEncryptioni ciphertext, they have been employed as mentioned above. In the same

way, the random noisy Salsa20, it was implemented and can be computed as follows: RandomNoisySalsai =

Char(Ord(StandardSalsaEncryptioni)+Ord(Ki))&Char(Ki)&Char(StandardSalsaEncryptioni) (m´od 120).

Where: The +operator refers to the sum function, while the &operator corresponds to the concatenation fun-

ction. The Ki shifting vector, it contains random integer values. The StandardEncryptioni parameter, it refers

to the ciphertext obtained using a standard encryption algorithm (e.g., DES, 3DES, Blowﬁsh, and AES-256),

as described in [5]. In a similar vein, the StandardSalsaEncryptioni argument refers to the ciphertext ob-

tained through the Salsa20 algorithm. Each strategy was applied separately. The RandomNoisyi sequence is

the F inalP ackage obtained through a random noisy strategy, as per [5]. In distinction, RandomN oisySalsai

denotes the F inalP ackage that results from employing random noisy Salsa20. The Ord function returns the

ordinal value of a character or integer, and the Char function converts a value to its corresponding character

representation in ASCII or UTF-8. In the random noisy proposals, a standard encryption algorithm is ﬁrst

used to encrypt the plaintext, and then random Caesar II mod 120 is applied to the ciphertext, adding an extra

layer of security [5–7]. In [5], it emphasizes that this approach diﬀers from double encryption and warns against

the repeated application of standard encryption algorithms, which could be exploited by cybercriminals using

computational strategies to decrypt the data. As noted in [5], noisy injection should be applied selectively to

the ciphertext to prevent arousing suspicion about the noise’s location. By applying it to only a portion of the

ciphertext, cybercriminals would encounter the daunting task of identifying the location of noisy characters, a

challenge that remains substantial even with the advent of quantum computing. These strategies involve the

use of random Caesar II mod 120 [5], applied to ciphertext previously obtained through a standard encryption

algorithm. Hence, they are considered as dynamic encryption alternatives for random performance [4]. Utili-

zing random noisy strategies for encryption has resulted in dynamic ciphertext, which plays a crucial role in

strengthening data security in organizational contexts. The random Caesar II methodology with (m´od 120) is

also an AI-based technique, leveraging random and heuristic methods to select the optimal Ki shift vector [5].

Thus, artiﬁcial intelligence was utilized when the heuristic method was applied to select the Ki vector that

yields maximum values for the encryption alphabet. The similarity in procedures between the heuristic method

and genetic algorithms leads to the consideration of AI application. This heuristic approach is employed to

validate the selected ASCII characters [5,8], with further explanation provided in [8]. A genetic algorithm (GA)

is a random procedure comprising selection, crossover, and mutation phases, followed by an evaluation stage

utilizing a wrapper [9,10] or ﬁtness function [10,16] for each GA generation [1,2,9,10,16,56,59–62]. In these

terms, random Caesar methodology [4,5], only selection procedure of GA is employed for selecting the alphabets

with (m´od 120) and its Ki vector of shifts. In this case, the range of Ki values are delimited between 30 and

150, for it does not exceed the 255 ASCII value. Similarly, the reduced random Caesar strategy [4,5], can uti-

lize the GA selection procedure with (m´od 120). For reduced random mutation [4], the ﬁrst phase (selection)

Facultad de Ingeniería

Universidad La Salle, Arequipa, Perú

facin.innosoft@ulasalle.edu.pe

143

Revista Innovación y Software

Vol. 6, No. 2, Mes Septiembre - Febrero, 2025

ISSN: 2708-0935

Pág. 134-158

https://revistas.ulasalle.edu.pe/innosoft

and third stage (mutation) of the GA model are employed. In both cases, the Ki shifting range is limited to

ordinal values between 0 and 105 to avoid exceeding the ASCII table maximum. This research did not cover

reduced random Caesar and reduced random mutation, but they could be explored in future work. Regarding

result evaluation, a modiﬁed cross-validation method [4–7], it was proposed to internally bias the discrimination

process, building on previous discussions. This information can help us recommend the employment of noisy

injection as a safety measure in organisations.

Results and Discussion

All experiments were carried out using T S, as previously mentioned. Thereby, the estimated error, T C, and

T D, they were computed separately for each encryption strategy, as described above. A ﬁve-fold modiﬁed cross-

validation [4], it was employed for each T S, allowing for direct comparison with the results of other studies [5–7].

According to [7], some studies employing the traditional cross-validation method [46,47,56,59,63,65–67] divide

the T S into ﬁve subsamples, each containing approximately 20 % of the data, with a similar number of elements

per class. One subsample is then used as the test sample (MC) to evaluate the model. The remaining four

subsamples (roughly 80 % of T S), they are joined together to form a training set, which is used to train the

model. The model’s performance is then evaluated using the MC as if it were new, unseen patterns. To obtain

the standard deviation after calculating the average or global accuracy, the process is repeated ﬁve times [66].

This traditional procedure is not applicable to data encryption/decryption, as the evaluation of the encryption

algorithm in this research does not require a training model with T S or evaluation with MC. Consequently,

the training and evaluation tasks are carried out before the ciphertext or F inalP ackage is generated. Initially,

the standard encryption algorithm is used to encrypt the plaintext, producing a ciphertext. This ciphertext is

then decrypted, and both the original and decrypted vectors are stored in the T S. For random noisy strategies,

the standard encryption algorithm is applied to the plaintext to generate a ciphertext. Then, the heuristic

method is used to simulate a partial phase training, utilizing only the random selection stage of the GA, as

applicable. The Ki vector is obtained by partial training using the random Caesar strategy, which it is applied

to the ciphertext for noisy injection. The encrypted sequence is decrypted, and both vectors are stored in the

T S. The novel cross-validation modiﬁcation [4,5], it has been updated to adopt a new approach that does

not rely on MC for assessing global accuracy [66]. The data encryption/decryption process in this research

uses a modiﬁed cross-validation approach [4–7], which it omits MC pattern evaluation. The error is computed

using four subsamples of the T S, and this process is repeated ﬁve times, with about 20 % of the T S being

sequentially omitted. The schema simulates the estimated error in diﬀerent environments by omitting a part

of the T S, which it ensures a more convergent result with an optimistic bias (i.e., the obtained value can be

better or equal in practical applications) [7]. This approach allows for the assessment of the estimated error

and other numerical features of the T S. If the decrypted ciphertext does not match the original plaintext, the

Facultad de Ingeniería

Universidad La Salle, Arequipa, Perú

facin.innosoft@ulasalle.edu.pe

144

Revista Innovación y Software

Vol. 6, No. 2, Mes Septiembre - Febrero, 2025

ISSN: 2708-0935

Pág. 134-158

https://revistas.ulasalle.edu.pe/innosoft

error percentage is determined by calculating the proportion of coded characters that they were not decrypted

correctly. Similarly, as the traditional cross-validation does, the operation is repeated ﬁve times, extracting

sequentially, a diﬀerent subsample in each iteration. With purpose of calculating the average and standard

deviation of each attribute or column of numeric type, which in this research, it was applied to the encryption

times (T C), decryption times (T D), and error percentage, globally, without distinguishing the elements by

class. As encryption ambiguity was noted during the experimentation process, the cross-validation was perfor-

med without distinguishing between classes. This topic may be revisited in future research to provide more

comprehensive explanations. However, this situation has not aﬀected the global accuracy of the encryption

strategies here evaluated. As a result, the experimentation in this work was limited to two classes of situations.

First, the comparison of ﬁve standard encryption algorithms (DES, 3DES, AES-256, Blowﬁsh, and Salsa20),

they were studied being applied to plaintext as static encryption schema for results comparisons with other

authors [5–7]. Secondly, we evaluated four random noisy strategies from [5], using ciphertext, as previously

mentioned. Speciﬁcally, we studied random noisy DES, random noisy 3DES, random noisy AES-256, and ran-

dom noisy Blowﬁsh, while the ﬁfth strategy, random noisy Salsa20, it is introduced here as a novel alternative.

Therefore, we experimented with the ﬁve random noisy strategies separately to inject noise into the ciphertext

as a dynamic encryption approach. After processing all the samples for each encryption strategy separately,

the global average results, they were computed using the novel updated cross-validation method [4–7], as ex-

plained above. This information can be found in Table 1, which includes the standard deviation in parentheses.

For comparison with other research [5,7], in columns T C and T D are shown the encryption and decryption

times respectively, which they were measured in milliseconds. The iterative procedure was stopped in this

study after ﬁve repetitions of ciphertext were generated for each encryption method. The results in Table 1,

which they include the standard deviation in parentheses, they are based on the average of ﬁve sequential

experiments here evaluated using the updated cross-validation method [5–11]. The parameters used for the

standard encryption methods, they are described below: The DES algorithm was conﬁgured with a 56-bit se-

cret key (’00000001’), UTF-8 encoding, and ECB mode [20], with ciphertext output in hexadecimal format.

The algorithm’s development leveraged the pycryptodome library from Python’s language [74,75]. To obtain

ciphertext with the TripleDES (3DES) algorithm, the ECB format [20], and OpenSSL [21,34] scheme, they

were used with a 24-bit secret key (’000000000000000000000001’), resulting in hexadecimal-encoded cipher-

text. The 3DES algorithm was implemented in Python using the cryptography package [73]. In the Blowﬁsh

algorithm, ECB format [20], with OpenSSL-based default_backend() and a 16-bit secret key (’00000001’),

they were employed, producing hexadecimal-encoded ciphertext. The implementation was also carried out

using Python’s Cipher module from cryptography package [73]. For AES-256 experiments, a 256-bit secret

key (’00000000000000000000000000000001’) and a 128-bit IV vector (’0000000000000001’), they were utilized.

The implementation involved CBC mode [20], with OpenSSL [21,34], and PKCS7 padding [21,34], with 128

bits, generating ciphertext in hexadecimal format. Python’s implementation leveraged the cryptography packa-

Facultad de Ingeniería

Universidad La Salle, Arequipa, Perú

facin.innosoft@ulasalle.edu.pe

145

Revista Innovación y Software

Vol. 6, No. 2, Mes Septiembre - Febrero, 2025

ISSN: 2708-0935

Pág. 134-158

https://revistas.ulasalle.edu.pe/innosoft

ge [73]. For the Salsa20 parameters, a 32-bit secret key with the value ’00000000000000000000000000000001’was

used, along with a nonce value of ’00000001ánd a mode operation of Ñone’. The pycryptodome library [74,75],

it was deﬁned the other requirements using standard values, and the ciphertext outputs were obtained with

hexadecimal encoding. The calculation of encryption/decryption times (in milliseconds) and estimated error,

they are shown in Table 1. Some of tgese values were rounded to match the results presented in [5,7]. Except

for ciphertext with the random noisy strategies because these processes of the encryption method are random.

Hence, it always generates diﬀerent and dynamical results. The columns T C and T D, they are the average

times of the estimated procedure for encryption and decryption tasks, respectively (the standard deviation

among parentheses is shown). Two tests of ciphertext results for each encryption strategy, they are also shown.

Two experimental tests with the same plaintext: "#Welcome#?", in some cases, diﬀerent results have been ob-

tained. In contrast, the majority of the experiments yielded successful results. With the exception of some tests

using the DES algorithm and its associated random noisy scheme, which they have reported errors. Characters

beyond the ASCII range, including ’#’(ordinal 9619) and ’?’(ordinal 65533), introduced into the plaintext, it

could be the cause.

Encryption

Strategies

TC TD %

Error

Test 1 Test 2

Salsa20 1.0030

(0.4828)

0.0882

(0.0058)

0 (0) c35d5f4be4-

3284e385d7-

77345fdf61d243

c35d5f4be4-

3284e385d7-

77345fdf61d243

random noisy

Salsa20

1.0232

(0.3491)

0.2717

(0.0311)

0 (0) ±N±ººH6{F{-

4ÇÇ*©D©±±-

¿¿Ps;s··Û-

vÛW$W³{³q-

<qÖrÖ^l5lÉÉ¦s¦Å-

Åj5j*ÇcÇææ··h7hä-

äl:l[T

Â_ÂzGzhÌh-

Ì±|±Á[Á\ãã7Xo-

<oo=o»»··åån;nk3k-

]2P¼¼ÂÂU-

c/c¯z¯ÊdÊ-

:àzà¬v¬S-

£?£QQ_+_ÁÁ

DES 1.1439

(0.5087)

0.1937

(0.0187)

1 (0) d535c337be-

1d94db5b94-

0b75bce124-

dcda93ﬀa35791bf07

d535c337be-

1d94db5b94-

0b75bce124-

dcda93ﬀa35791bf07z

Facultad de Ingeniería

Universidad La Salle, Arequipa, Perú

facin.innosoft@ulasalle.edu.pe

146

Revista Innovación y Software

Vol. 6, No. 2, Mes Septiembre - Febrero, 2025

ISSN: 2708-0935

Pág. 134-158

https://revistas.ulasalle.edu.pe/innosoft

Encryption

Strategies

TC TD %

Error

Test 1 Test 2

random noisy

DES

1.6333

(0.4897)

0.5848

(0.0389)

1 (0) ¿[¿²}²c0c¹¹à}ài6iÈ-

ÈÀÀíí²M²-

l;lUM0ºXº-

j5j£A£½½gj:j=Ui4i¥-

C¥(«F«ÁÁly-

ññÐmÐÐlÐôôi0i-

a.a°J°&7[-

n9na*a»»O-

O7÷÷\,\i2i

“ððÉÉ]Y¾[¾\)\

m S¨F¨ÆaÆg-

6gññM¬x¬-

óóðð§r§ÃÃ«w«]«I«-

Wr=r¿]¿¯-

L¯ôôV %V{-

I{¼¼¸T¸åå2=ººRRÑ-

kÑììôôdr=rY”YÆÆ-

|à~àßyßp@p®w®”

AES-256 7.3116

(3.7093)

0.6166

(0.0169)

0 (0) 3030303030-

3030303030-

3130319421-

780fc40c59-

f14796a598-

115d8e5139-

f6f9666f67-

cb30fcda01-

f22739eeda

3030303030-

3130319421-

780fc40c59-

f14796a598-

115d8e5139-

f6f9666f67-

cb30fcda01-

f22739eeda

Facultad de Ingeniería

Universidad La Salle, Arequipa, Perú

facin.innosoft@ulasalle.edu.pe

147

Revista Innovación y Software

Vol. 6, No. 2, Mes Septiembre - Febrero, 2025

ISSN: 2708-0935

Pág. 134-158

https://revistas.ulasalle.edu.pe/innosoft

Encryption

Strategies

TC TD %

Error

Test 1 Test 2

random noisy

AES-256

8.4637

(3.9351)

1.0021

(0.0119)

0 (0) xExdj7jµµY-

&Y¡q¡eÄÄ¼-

¼W’Wk8ki9-

iU¨x¨p=p¾-

¾j7jZµµ^.-

^iWOQ!QV-

#Vej7jR³³-

{K{OUY§s§uCu«z«

i ÍÍ®~®ÍgÍ-

ËhËs?svFv-

÷÷{F{ÍÍ¯-

I¯¸¸h£l£r-

9rZ¿^¿ÄÄ-

³z³¤l¤R^¯-

z¯$ÌÌá|á²}²½½}J}È-

ÈÈbÈ°z°¤-

>¤¼¼h2hfq-

;q¦@¦ËËS£-

@£´R´ÀÀÄ-

Äõõââ¥A¥-

&[h7h#¸¸µ-

µ²{²j7jn5nðð$

¥D¥

“§t§P P[(-

[xHxNuEuÇ-

ÇnÁÁ´´Ox-

Hx»»¹¹µµ-

d¬y¬]ÁÁ_-

/_n;nÀÀMS-

#S‘W’Wo<o-

xHxªwª\&\

V¯~¯v=vÅ-

Å¥s¥»»ÀÀ-

p8pT²L²íí-

£o£c3cµRµ-

¼¼_&_Ä^Ä}-

L}Z¶¶fTTí-

í··G~F~V %-

VN¢m¢÷÷¾-

¾9u@uÄÄ]*-

]\ÕoÕKÔn-

ÔdQzDz¸¸-

â|âX”Xh½Z½ÍkÍ¿¿h-

*<4Á‘ÁVk5U#U

n [$[U”UXÉdÉõ-

õÔpÔÀ_À”

Blowﬁsh 7.5597

(3.7931)

0.6486

(0.0373)

0 (0) 4af747eaab-

e473251f42-

200cf8fda7f2

4af747eaab-

e473251f42-

200cf8fda7f2

Facultad de Ingeniería

Universidad La Salle, Arequipa, Perú

facin.innosoft@ulasalle.edu.pe

148

Revista Innovación y Software

Vol. 6, No. 2, Mes Septiembre - Febrero, 2025

ISSN: 2708-0935

Pág. 134-158

https://revistas.ulasalle.edu.pe/innosoft

Encryption

Strategies

TC TD %

Error

Test 1 Test 2

random noisy

Blowﬁsh

7.8544

(3.7619)

0.9457

(0.0349)

0 (0) ÉÉïïóóc,c¦-

r¦_3óóÖuÖ-

ÇeÇÇbÇk¨q-

¨hiY$Y_±K-

±bY’Y½½³³-

¹¹ÑnÑ¨B¨H-

ÐjÐ»W»èè®-

w®±K±¦t¦

YÓrÓ5‘UcÁ-

\ÁÃbÃ®M®-

Á_Áéé[’[-

m6mi6ic1c£-

n£½½µOµ©u-

©³³[)[±±b-

×t×Â\Ât<t-

.à|à¥D¥ÂÂ.]

3DES 7.6814

(3.9388)

0.5004

(0.0184)

0 (0) d535c337be-

1d94db5b94-

0b75bce124-

dc620c51e2380a3d5c

d535c337be-

1d94db5b94-

0b75bce124-

dc620c51e2380a3d5c

random noisy

3DES

8.4594

(4.1892)

0.6742

(0.0179)

0 (0) êêTTp=pµµ.-

e‘ªsªÈfÈÄ_Äk:k

<§n§ÉÉµQµ-V*-

y@yS[¤B¤À-

ÀÂÂÖ°N°µRµ-

,d3dd2dRRË-

gËÂ_Â£m£Ã-

ÃT¦C¦m8mh7h ;

Z(Z m j2j p çç_-

,_ÓoÓ¡l¡öö

8xCxxEx^)^-

¤A¤dn;n[$-

[çç7®}®&w-

>wxDxéé³Q-

³µµºXº¢i¢-

³³²²ºXºÆÆ-

³~³+ÙvÙÓn-

Ó£r£[)[])]&¡>¡iY’-

Y´´ÊgÊ¡l¡-

[óóYU”U°x°j:jììXÂ^

Â¼¼3

Average 5.2134

(3.2975)

0.5526

(0.2849)

0.2

(0.40)

- - - - - - - -

While these characters might be unavoidable in real applications, in Table 1, it demonstrates that most encry-

ption methods, they were able to hide this issue in the encrypted output. Reference [5], it notes that a common

issue with random noisy alternatives is the inability to control the maximum random value selected from

the ASCII table. Despite the success of these strategies, an ASCII value can be reinterpreted as a diﬀerent

encoding character, such as UTF-8. By employing mod 120, which yields sequences with allowed ASCII va-

lues, we avoided these problems in our study. Unless in particular instances where plaintext input was subject

Facultad de Ingeniería

Universidad La Salle, Arequipa, Perú

facin.innosoft@ulasalle.edu.pe

149

Revista Innovación y Software

Vol. 6, No. 2, Mes Septiembre - Febrero, 2025

ISSN: 2708-0935

Pág. 134-158

https://revistas.ulasalle.edu.pe/innosoft

to noisy injection, as discussed above. Compared to other standard symmetric proposals, the Salsa20-based

encryption process, it was substantially faster, with speeds 1.1404 to 7.6577 times greater. This resulted in

a millisecond diﬀerence of 0.1408 to 6.6783 (see Table 1). Despite the encryption/decryption times, they are

much faster using Salsa20 algorithm, in this research, it was observed that this strategy supports plaintext

or ciphertext with values greater than 255 characters. Thereby, it can be considered a secure schema if this

situation is validated properly. Similarly, the 3DES alternative has not encountered any errors, but it is li-

mited to supporting a maximum of 22 characters for both plaintext and ciphertext, similar to the Blowﬁsh

proposals with 13 maximum. In this research, DES has a character limit of 255 for plaintext or ciphertext.

Given the average error rate of 1.0 % during data processing, the DES alternative, it is not considered a reliable

option. Furthermore, Table 1 in this study presents several static ciphertext results obtained using standard

encryption algorithms such as Salsa20, DES, 3DES, AES-256, and Blowﬁsh. However, this does not necessarily

imply that these schemes are vulnerable or insecure in all cases. Moreover, random Caesar applied to plaintext

achieved the best balance. Notably, the random Caesar II with mod 120 signiﬁcantly improved encryption

times, outperforming the other strategies evaluated. The experiments yielded an average encryption time of

0.14 milliseconds with a standard deviation of 0.0108. In contrast, the decryption process had an average time

of 0.05 milliseconds with a standard deviation of 0.0011. These results are not included in Table 1, as the focus

of this research is on comparing standard encryption algorithms with their noisy injection applications. When

applied to ciphertext, the random noisy Salsa20 strategy, which it combines standard Salsa20 with random

Caesar II mod 120, outperformed the rest random noisy approaches here evaluated. The most balanced results

are consistently achieved with Salsa20 strategy that it includes its random noisy strategy. Compared to the

rest random noisy strategies, the novel random noisy Salsa20 alternative has showed a notable performance

improvement, with speeds 1.5962 to 8.2712 times faster. The diﬀerence in milliseconds ranged from 0.6101 to

7.4404. The results show that the encryption speed diﬀerence between Salsa20 and its random noisy version, it

was not signiﬁcant. For instance, traditional Salsa20 is 1.0201 times faster, with a marginal diﬀerence of 0.0201

milliseconds (see Table 1). In this study, both Salsa20 alternatives, they were found to support a maximum

plaintext length of 255 characters, as previously noted. Notably, the experiments did not reveal any errors

related to this limitation. At any rate, this measure is deemed to provide a substantial improvement in the

conﬁdence of the encryption strategy’s performance. The random noisy schemes demonstrate potential as an

experimental tool, yet further investigation into other aspects is necessary, particularly since the experiments,

they were restricted to plaintext inputs of up to 255 characters. The cases of 3DES, and Blowﬁsh are exceptions,

as previously discussed. Each encryption strategy was evaluated based on its own training sample, which was

designed independently. Random Caesar schemes have proven eﬀective in enhancing data security [4,5,9,11],

and the random noisy strategies, they have demonstrated similar eﬃcacy. Table 1 conﬁrms this eﬀect in the

global average calculation. Notably, random noisy alternatives result in slightly longer ciphertexts. The experi-

ments revealed that DES-generated ciphertext can be quicker than 3DES, AES-256, and Blowﬁsh. Nonetheless,

Facultad de Ingeniería

Universidad La Salle, Arequipa, Perú

facin.innosoft@ulasalle.edu.pe

150

Revista Innovación y Software

Vol. 6, No. 2, Mes Septiembre - Febrero, 2025

ISSN: 2708-0935

Pág. 134-158

https://revistas.ulasalle.edu.pe/innosoft

the use of DES may pose security risks due to its potential vulnerability to decryption, as evidenced by the

1.0 % average error rate observed during encryption and decryption. The use of random noisy strategies with

standard encryption algorithms produced dynamic ciphertext results in every instance. Nonetheless, the time

required for FinalPackage ciphertext generation exceeded that of the standard algorithms, as demonstrated

in Table 1’s Test1 and Test2. The execution times for Salsa20 proposals, they are shorter compared to the

DES, 3DES, AES-256, and Blowﬁsh strategies here evaluated. Notably, these random noisy schemes consis-

tently produce dynamic ciphertext outputs. Cybercriminals would encounter signiﬁcant obstacles in decrypting

data, as they would need to determine each random Ki shifting value in advance, which has been previously

hidden. When utilizing the PartialNoisy proposal, a novel partial noisy injection scheme introduced in [5], the

decryption process becomes notably more complex. The complexity of these data discovery tasks could make

them diﬃcult to accomplish, even with quantum computing. The exploration of PartialNoisy was beyond the

scope of this study and is suggested for future work. Repeatedly using these random noisy strategies, they can

lead to better and more dynamic results in comparison to traditional static encryption methods. This noisy

injection alternative, it can increase the security degree of the ciphertext or plaintext. Besides, this situation

might warn us against future quantum computing attacks [8,22,23,42], improving the digital data security of

the organizations, as mentioned above. Additionally, the resilience of these random noisy alternatives to various

cyberattacks remains unevaluated, leaving potential vulnerabilities unknown. Comparing to other studies [4,5],

our research suggests that downsized ciphertext using reduced random or mutation methodologies [4], which

they can be eﬀective indicators. These schemes provide a balance between eﬃciency and security, yielding short

ciphertexts and fast encryption processes, while the Ki shifting of partial ciphertext ensures the protection of

data security. These alternatives can mitigate the risk of digital data theft by incorporating a signiﬁcant amount

of noise into the ciphertext. In other words, proposals based on reduced random and mutation schemes show

promise. However, this study focuses on other approaches, leaving reduced and mutation alternatives for future

works. Notwithstanding the diﬃculties, the study’s goals and hypotheses, they were fulﬁlled successfully. Due to

that the random noisy Salsa20, it has been presented here as a novel proposal based on noisy injection because

it allows obtaining dynamic encryption ciphertext outputs. In other words, it can generate varying outcomes,

even with identical plaintext input and parameters, thereby confusing potential cybercriminals. This data is

supported by the information presented in Table 1. By comparing standard encryption algorithms with random

noisy strategies, we can infer that noisy injection is a promising approach for organizations seeking a safe alter-

native. This approach incorporates the novel random noisy Salsa20 strategy, especially in environments where

traditional Salsa20 methods, they are already in use. Implementing the random noisy scheme is recommended

to strengthen digital data security in these cryptosystems. This novel alternative based on noisy injection is

considered a safe measure for organizations, as it opens up a wide range of opportunities for improving digi-

tal data security. Furthermore, a modiﬁcation to the dynamic encryption methodology presented here could

be worth exploring. By utilizing reduced noisy strategies [5], and reduced random mutation schemes [4], this

Facultad de Ingeniería

Universidad La Salle, Arequipa, Perú

facin.innosoft@ulasalle.edu.pe

151

Revista Innovación y Software

Vol. 6, No. 2, Mes Septiembre - Febrero, 2025

ISSN: 2708-0935

Pág. 134-158

https://revistas.ulasalle.edu.pe/innosoft

approach enables a signiﬁcant reduction in ciphertext size, up to 33 %, in FinalPackage compared to random

noisy proposals. Another measure that it might be explored, it is the application of diﬀerent ways for achieving

the noisy injection. Especially, those that merge the simultaneous random noisy methodology with artiﬁcial

intelligence utilizing the nearest neighbor rule [8,10,11,46,47,65,67], and pseudo-hexadecimal encoding [8–10],

as noted above. This implies a vast array of possibilities that could be explored in future research.

Conclusions

Regular updates to cybersecurity strategies, including encryption methods, they are vital for maintaining

digital data safety in organizations. However, it does not guarantee their digital data security. Studies cited

above [4,8], they have revealed that the challenges posed by inadequate encryption measures. If a method is

well-known, it will become in inadequate strategy. Previous studies [4–11], they have proposed various dynamic

encryption alternatives to address the issue of digital data theft. We present a novel approach that fuses

standard encryption techniques with a random Caesar strategy, tailored for real-world use in organizations. A

new dynamical encryption proposal, known as the random noisy Salsa20 strategy, is presented in this paper.

Furthermore, the evaluation included a comparison of ﬁve dynamic encryption alternatives that utilize random

noisy strategies. These strategies leverage artiﬁcial intelligence for noisy injection into ciphertext, utilizing

random and heuristic methods as previously discussed. Therefore, it has the potential for practical application in

organizational contexts. The methodology based on noisy injection makes a valuable contribution to overcoming

the limitations of standard encryption strategies, which in turn boosts its eﬀectiveness. Experimental results

with the dynamic encryption random noisy alternatives, they have revealed that can cope with the cyberattacks

and data security problems with high levels of trust. Hence, despite of the occurred diﬃculties, the research’s

hypotheses and objectives, they were fulﬁlled successfully. Due to that the random noisy Salsa20, it has been

presented here as a novel proposal based on noisy injection because it allows obtaining dynamic encryption

ciphertext outputs. Besides, in all cases, these random noisy strategies, they allow the dynamic generalized

results even better than those obtained with the standard encryption algorithms (see Table 1). We plan to

conduct further investigation into this matter. One of the methods we’ll be exploring is the application of

measures to decrease the size of ciphertext resulting from random noisy strategies. The use of reduced random

schemes [5], or reduced mutation strategies [4], it might be an option, since they enable the masking of ciphertext

operations. Another potential approach to consider is the use of various methods for achieving noisy injection.

In particular, the methods that synergize random noisy approaches with nearest neighbor rule-based AI and

pseudo-hexadecimal encoding, as noted earlier. As expected, this involves a multitude of possibilities that we

can cover in future works.

Facultad de Ingeniería

Universidad La Salle, Arequipa, Perú

facin.innosoft@ulasalle.edu.pe

152

Revista Innovación y Software

Vol. 6, No. 2, Mes Septiembre - Febrero, 2025

ISSN: 2708-0935

Pág. 134-158

https://revistas.ulasalle.edu.pe/innosoft

Acknowledgment

This work was supported by Tecnológico Nacional de México (I.T. Ciudad Altamirano). This research is future

work of a last project identiﬁed by: 19329.24-P.

Contribución de Autoría

Edgar Rangel Lugo: Administración de proyectos,Investigación,Metodología,Software,Validación (random

noisy strategies), Redacción - borrador original. Kevin Uriel Rangel Ríos: Conceptualización,Investigación,

Software,Validación (several experiments, random noisy strategies), Escritura, revisión y edición (reviewer,

editing, and translation). Carlos Alberto Bernal Beltrán: Investigación,Validación (standard encryption al-

gorithms). Leonel González Vidales: Investigación,Validación (standard encryption algorithms). César Del

Ángel Rodríguez Torres: Investigación. Lucero De Jesús Ascencio Antúnez: Validación (standard encryption

algorithms). Rosa Isabel Reynoso Andrés: Validación (standard encryption algorithms).

Referencias

[1] B. Delman, “Genetic algorithms in cryptography,” Master’s thesis, Rochester Institute of Technology, 2004,

rIT Scholar Works.

[2] S. Kalsi, H. Kaur, and V. Chang, “Dna cryptography and deep learning using genetic algorithm with nw

algorithm for key generation,” Journal of Medical Systems, vol. 42, no. 17, 2018.

[3] J. C. Mendoza, “Demostración de cifrado simetrico y asimétrico,” Ingenius, Revista de Ciencia y Tecnología,

no. 3, pp. 46–53, 2008.

[4] E. Rangel and K. U. Rangel, “Novel random encryption methods based on mutation strategies of artiﬁcial

intelligence,” SPCSJ: Scientiﬁc and Practical Cyber Security Journal, vol. 8, no. 3, pp. 84–91, 2024.

[5] E. Rangel, K. U. Rangel, and L. González, “Dynamic encryption methods based on noisy injection and ca-

mouﬂaging ciphertext strategies with artiﬁcial intelligence,” SPCSJ, Scientiﬁc and Practical Cyber Security

Journal, vol. 9, no. 1, pp. 82–104, 2025.

[6] ——, “Inyección de ruido para encriptado de datos dinámico con inteligencia artiﬁcial. caso de estudio:

Algoritmo gost r 34.12-2015,” Revista Electrónica de Divulgación de la Investigación del SABES, vol. 29,

2025.

Facultad de Ingeniería

Universidad La Salle, Arequipa, Perú

facin.innosoft@ulasalle.edu.pe

153

Revista Innovación y Software

Vol. 6, No. 2, Mes Septiembre - Febrero, 2025

ISSN: 2708-0935

Pág. 134-158

https://revistas.ulasalle.edu.pe/innosoft

[7] E. Rangel and K. U. Rangel, “Mejorando la seguridad del algoritmo camellia, mediante la inyección de

ruido sobre textos cifrados utilizando procesos basados en inteligencia artiﬁcial,” INTELETICA, Revista

de Inteligencia Artiﬁcial, Ética y Sociedad, vol. 2, no. 3, 2025.

[8] E. Rangel, K. U. Rangel, L. González, A. Ortiz, and C. A. Rodríguez, “Four dynamic encryption alternatives

with artiﬁcial intelligence based on pseudo-hexadecimal noisy injection schema for handling the theft of

digital data problem,” SPCSJ, Scientiﬁc and Practical Cyber Security Journal, vol. 9, no. 3, pp. 59–77,

2025.

[9] E. Rangel, K. U. Rangel, J. Medrano, C. A. Bernal, and L. González, “Algoritmo genético para cifrado de

datos, basado en un nuevo concepto pseudo-hexadecimal con inteligencia artiﬁcial,” in VI Congreso Na-

cional De Investigación En Ciencia E Innovación De Tecnologías Productivas, Cd. Altamirano, Guerrero,

México, 2023.

[10] E. Rangel, K. U. Rangel, and L. González, “Cifrado de datos dinámico con inteligencia artiﬁcial, utilizando

el nuevo formato pseudo-hexadecimal,” Revista Electrónica de Divulgación de la Investigación del SABES,

vol. 28, 2024.

[11] E. Rangel and K. U. Rangel, “La regla del vecino más cercano como alternativa para inyectar ruido a men-

sajes encriptados por el algoritmo: Noised random hexadecimal,” INTELETICA, Revista de Inteligencia

Artiﬁcial, Ética y Sociedad, vol. 1, no. 2, pp. 1–15, 2024.

[12] D. Álvarez, “Algunos aspectos jurídicos del cifrado de comunicaciones,” Derecho PUCP, no. 83, pp. 241–

264, 2019.

[13] F. Barranco and C. Galindo, Criptografía básica y algunas aplicaciones. Universidad Jaume I, 2022.

[14] S. Gómez, J. D. Arias, and D. Agudelo, “Cripto-análisis sobre métodos clásicos de cifrado,” Scientia Et

Technica, vol. 2, no. 50, pp. 97–102, 2012.

[15] B. Javidi and J. L. Horner, “Optical pattern recognition for validation and security veriﬁcation,” Optical

Engineering, vol. 33, no. 6, pp. 1752–1756, 1994.

[16] B. Reddaiah, “A study on genetic algorithms for cryptography,” International Journal of Computer Ap-

plications, vol. 177, no. 28, pp. 1–4, 2019.

[17] C. Sebas, “¿qué son los algoritmos genéticos en las inteligencias artiﬁciales?” 2023.

[18] S. Paul, P. Dasgupta, P. K. Naskar, and A. Chaudhuri, “Secured image encryption scheme based on dna

encoding and chaotic map,” Review Of Computer Engineering Studies, vol. 4, no. 2, pp. 70–75, 2017.

Facultad de Ingeniería

Universidad La Salle, Arequipa, Perú

facin.innosoft@ulasalle.edu.pe

154

Revista Innovación y Software

Vol. 6, No. 2, Mes Septiembre - Febrero, 2025

ISSN: 2708-0935

Pág. 134-158

https://revistas.ulasalle.edu.pe/innosoft

[19] R. Oppliger, Contemporary cryptography. Artech House Computer Security Library, 2005.

[20] D. R. Stinson and M. B. Paterson, Cryptography: Theory and Practice, 4th ed. Chapman and Hall/CRC,

2019.

[21] H. C. A. Van-Tilborg, Encyclopedia Of Cryptography And Security. Springer, 2005.

[22] L. Baklaga, “Leading the way in quantum-resistant cryptography for everyday safety,” SPCSJ, Scientiﬁc

and Practical Cyber Security Journal, vol. 8, no. 3, pp. 65–73, 2024.

[23] R. Bavdekar, C. Eashan-Jayant, A. Ankit, and K. Tiwari, “Post quantum cryptography: A review of

techniques, challenges, and standardizations,” in International Conference on Information Networking

(ICOIN), 2023.

[24] Q. H. Dang and H. Q. Le, “Improved cryptanalysis of the rsa algorithm using side-channel attacks,” Journal

of Information Security and Applications, vol. 65, 2022.

[25] D. Luciano and G. Prichett, “Cryptology: From caesar ciphers to public-key cryptosystems,” The College

Mathematics Journal, vol. 18, pp. 2–17, 1987.

[26] M. S. Rahman and M. S. Hossain, “A secure private key cryptography scheme using rsa and aes,” Journal

of Cybersecurity, vol. 1, no. 1, pp. 1–9, 2021.

[27] J. Rodríguez, Operadores Genéticos Aplicados A La Criptografía Simétrica. Universidad Distrital Fran-

cisco José De Caldas, 2020.

[28] M. Baker and J. Schiller, “Ecies: Elliptic curve integrated encryption scheme,” in Cryptography and Network

Security. Springer, 2015, pp. 245–263.

[29] D. Hankerson, J. L. Hernandez, and A. J. Menezes, “Software implementation of elliptic curve cryptography

over binary ﬁelds,” in CHES 2000, LNCS 1965. Springer-Verlag, 2004, pp. 1–24.

[30] P. L. Montgomery, “Speeding up the pollard rho method,” Mathematics of Computation, vol. 48, no. 177,

pp. 453–456, 1987.

[31] NIST, “Recommended methods for key establishment using public key cryptography,” NIST Special Pu-

blication 800-56A, Tech. Rep., 2013.

[32] H. W. Dhany, F. Izhari, H. Fahmi, M. Tulus, and M. Sutarman, “Encryption and decryption using password

based encryption, md5, and des,” Atlantis Press, 2018.

Facultad de Ingeniería

Universidad La Salle, Arequipa, Perú

facin.innosoft@ulasalle.edu.pe

155

Revista Innovación y Software

Vol. 6, No. 2, Mes Septiembre - Febrero, 2025

ISSN: 2708-0935

Pág. 134-158

https://revistas.ulasalle.edu.pe/innosoft

[33] A. Kumar and S. Sharma, “A study on historical cryptographic techniques: Caesar cipher to des,” Inter-

national Journal of Advanced Science and Technology, vol. 30, no. 2, pp. 555–564, 2021.

[34] H. C. Van and S. Jajodia, Encyclopedia Of Cryptography And Security. Springer Science & Business

Media, 2011.

[35] B. Schneier, “Description of a new variable-length key, 64-bit block cipher (blowﬁsh),” in Fast Software

Encryption, 1994.

[36] ——, Secrets and lies: Digital security in a networked world. Wiley, 2000.

[37] E. A. AL-Maqtari, “Performance evaluation for aes, blowﬁsh, des, and 3des cryptography algorithms,”

PUIRP: Partners Universal Innovative Research Publication, vol. 2, no. 5, pp. 86–95, 2024.

[38] R. K. Muhammed, R. R. Aziz, A. A. Hassan, A. M. Aladdin, S. J. Saydah, T. A. Rashid, and B. A. Hassan,

“Comparative analysis of aes, blowﬁsh, twoﬁsh, salsa20, and chacha20 for image encryption,” Kurdistan

Journal of Applied Research, vol. 9, no. 1, pp. 52–65, 2024.

[39] H. K. Garai and S. Dey, “A multi-step key recovery attack on reduced round salsa and chacha,” Cryptologia,

vol. 49, no. 3, pp. 252–267, 2024.

[40] A. Saini, A. Tsokanos, and R. Kirner, “Cryptoqnrg: a new framework for evaluation of cryptographic

strength,” The Journal of Supercomputing, vol. 79, pp. 12 219–12 237, 2023.

[41] J. Daemen and V. Rijmen, The Design of Rijndael: AES - The Advanced Encryption Standard. Springer,

2002.

[42] M. Iavich, T. Kuchukhidze, and A. Gagnidze, “Post-quantum digital signature using verkle trees and

lattices,” SPCSJ, Scientiﬁc and Practical Cyber Security Journal, vol. 8, no. 3, pp. 35–52, 2024.

[43] P. Fuegner, “Are rsa and aes both at risk from the quantum threat?” 2024.

[44] S. Sengupta and S. Ghosh, “Quantum computing encryption threats: Why rsa and aes are at risk,” Journal

of Cryptographic Research, 2023.

[45] J. Thakur and N. Kumar, “Des, aes and blowﬁsh: Symmetric key cryptography algorithms simulation

based performance analysis,” International Journal of Emerging Technology and Advanced Engineering,

pp. 6–12, 2011.

[46] E. Rangel, “Vecinos envolventes para variantes de la regla del vecino más cercano,” Master’s thesis, Instituto

Tecnológico de Toluca, 2002.

Facultad de Ingeniería

Universidad La Salle, Arequipa, Perú

facin.innosoft@ulasalle.edu.pe

156

Revista Innovación y Software

Vol. 6, No. 2, Mes Septiembre - Febrero, 2025

ISSN: 2708-0935

Pág. 134-158

https://revistas.ulasalle.edu.pe/innosoft

[47] ——, “La regla de los k vecinos más cercanos (k-nn) basada en distancia de manhattan (city-block) para

mejorar la clasiﬁcación de patrones,” in V Congreso Nacional De Investigación En Ciencia E Innovación

De Tecnologías Productivas, 2022.

[48] S. S. Iyengar, R. Kannan, and S. Ganapathi, “Inteligencia artiﬁcial y criptografía: Tendencias y desafíos,”

IEEE Transactions on Emerging Topics in Computing, vol. 9, no. 2, pp. 833–844, 2021.

[49] X. Liu and X. Wang, “Quantum cryptanalysis of lattice-based cryptographic protocols,” Physical Review

X, vol. 12, no. 2, 2022.

[50] A. K. Singh, P. Kumar, and R. Singh, “Aplicación de la inteligencia artiﬁcial en la criptografía: Una

revisión,” Journal of Intelligent Information Systems, vol. 67, no. 2, pp. 257–275, 2021.

[51] T. M. Mitchell, Machine learning, 2nd ed. McGraw-Hill, 2020.

[52] J. Ross-Quinlan, C4.5: Programs for Machine Learning. Morgan Kaufmann, 1993.

[53] S. J. Russell and P. Norvig, Inteligencia artiﬁcial: Un enfoque moderno, 4th ed. Pearson, 2020.

[54] A. K. Hartmann, “Heuristic search in graphs,” Journal of Artiﬁcial Intelligence Research, vol. 68, pp. 1–33,

2020.

[55] J. S. Sánchez, F. Pla, and F. J. Ferri, “Prototype selection for the nearest neighbor rule through proximity

graphs,” Pattern Recognition Letters, vol. 18, pp. 507–513, 1997.

[56] L. I. Kuncheva and L. C. Jain, “Nearest neighbor classiﬁer: Simultaneous editing and feature selection,”

Pattern Recognition Letters, vol. 20, pp. 1149–1156, 1999.

[57] K. P. Murphy, Probabilistic machine learning: An introduction. MIT Press, 2022.

[58] B. Reddaiah, “A study on pairing functions for cryptography,” IJCA, vol. 149, no. 10, pp. 4–7, 2016.

[59] D. B. Skalak, “Prototype and feature selection by sampling and random mutation hill climbing algorithms,”

in Proceedings of the Eleventh International Conference on Machine Learning. Morgan Kaufmann, 1994,

pp. 293–301.

[60] A. Clark, “Modern optimisation algorithms for cryptanalysis,” in Proceedings of the 1994 Second Australian

and New Zealand Conference on Intelligent Information Systems, 1994, pp. 258–262.

[61] W. Griindlingh and J. H. Van-Vuuren, “Using genetic algorithms to break a simple cryptographic cipher,”

2002, submitted.

Facultad de Ingeniería

Universidad La Salle, Arequipa, Perú

facin.innosoft@ulasalle.edu.pe

157

Revista Innovación y Software

Vol. 6, No. 2, Mes Septiembre - Febrero, 2025

ISSN: 2708-0935

Pág. 134-158

https://revistas.ulasalle.edu.pe/innosoft

[62] R. A. J. Matthews, “The use of genetic algorithms in cryptanalysis,” Cryptologia, vol. 17, no. 4, pp. 187–201,

1993.

[63] L. Bruzzone and S. B. Serpico, “Classiﬁcation of imbalanced remote-sensing data by neural networks,”

Pattern Recognition Letters, 1997.

[64] I. Goodfellow, Y. Bengio, and A. Courville, Deep learning. MIT Press, 2021.

[65] R. Barandela, J. S. Sánchez, V. García, and E. Rangel, “Strategies for learning in class imbalance problems,”

Pattern Recognition, vol. 36, no. 3, pp. 849–851, 2003.

[66] D. Lewis and J. Catlett, “Heterogeneous uncertainty sampling for supervised learning,” in Proceedings of

the 11th International Conference on Machine Learning, 1994, pp. 148–156.

[67] T. M. Cover and P. E. Hart, “Nearest neighbor pattern classiﬁcation,” IEEE Transactions on Information

Theory, vol. 13, pp. 21–27, 1967.

[68] E. Rangel and K. U. Rangel, “Novel pseudo-hexadecimal encryption strategies for camouﬂaging ciphertext

based on nearest neighbor with artiﬁcial intelligence,” IJCOPI, 2024.

[69] Microsoft, “Descarga de software,” 2025.

[70] Python.org, “The python network,” 2024.

[71] Google, “Android 12, sistema operativo para dispositivos móviles,” 2024.

[72] Pydroid3, “Pydroid3 versión 7.4_arm64,” 2025.

[73] Python, “Cryptography 45.0.4,” 2025.

[74] PyCryptodome, “Crypto.cipher package,” 2025.

[75] PyPI, “Pycryptodome 3.21.0,” 2024.

Facultad de Ingeniería

Universidad La Salle, Arequipa, Perú

facin.innosoft@ulasalle.edu.pe

158