REVISTA INNOVACIÓN Y SOFTWARE VOL 4 Nº 2 Septiembre - Febrero 2023 ISSN Nº 2708-0935
RECIBIDO 25/03/2023 ● ACEPTADO 08/07/2023 ● PUBLICADO 30/09/2023
RESUMEN
Debido al creciente número de espectros inalámbricos, las múltiples frecuencias están enredando
el proceso de gestión de recursos, lo que dificulta el funcionamiento. Además, los datos anteriores
se vuelven vulnerables cuando se reciben informes de enigma de fuga de datos. En esta situación,
es indispensable asegurar los datos en el conjunto de datos y detectar la cantidad real de datos
durante el mecanismo de transformación de recursos en redes inalámbricas. Se ha desarrollado
un sistema para detectar la fuga de datos utilizando técnicas de prueba supervisadas y no
supervisadas mediante simulación en Python. Se obtienen los resultados previstos y reales, que
se reducen mediante pruebas supervisadas y no supervisadas, el resultado sigue siendo del
96,03% y 94,53% respectivamente.
Palabras claves:
Pruebas supervisadas, pruebas no supervisadas, redes neuronales, redes
inalámbricas.
ABSTRACT
Due to an increasing number of wireless spectrums, the multiple frequencies are tangling resource
management process that results hindrance in operation. In addition, the previous data become
vulnerable when reports are received for data leakage enigma. In this situation, it is indispensable
to secure the data in the dataset and detect the actual amount of data during resource
transformation mechanism in wireless networks. A system as been developed to detect the leaked
Zeeshan Aslam
Alfanar Global Development Saudi
Arabia. Damman, Arabia Saudita.
zeeshan.aslam@alfanar.com
nfc.ie@hotmail.com
ARK: ark:/42411/s12/a108
DOI: 10.48168/innosoft.s12.a108
PURL: 42411/s12/a108
REVISTA INNOVACIÓN Y SOFTWARE VOL 4 Nº 2 Septiembre - Febrero 2023 ISSN Nº 2708-0935
53
data using supervised and unsupervised testing technique by conducting simulation in Python.
The targeted and actual outcome is obtained which deduced through supervised and undersized
testing, the outcome remained 96.03%, and 94.53% respectively.
Keywords:
: Supervised testing, unsupervised testing, neural network, wireless networks.
INTRODUCTION
With the exponential increase in the number of wireless devices in recent years and due to the
rapid growth of wireless services, it has become increasingly important to allocate and manage
them accordingly. There is, however, generally a challenge to establish suitable strategies due to
the fact that the non-convex objective average rate maximization problem is NP-hard. It is no
secret that there has been a growing interest in numerical optimization as it relates to wireless
resource allocation in recent years.
There are still significant challenges in implementing numerical optimization-based algorithms on
practical systems, e.g., high computation costs, despite their ability to solve specific resource
management problems with tremendous results [1]
As neural networks (NNs) [2], memorize features of example data during training and
unintentionally reveal them during prediction, it is a major concern for machine learning
applications to prevent models from revealing sensitive input data details. There is, however, no
easy way to accomplish this. The literature is lacking studies that address deep learning-based
wireless resource allocation systems with privacy protection.
These radio resource management algorithms cannot always provide an adequate degree of
performance due to the dynamic nature of wireless networks. An algorithm that learns from
interactions with the environment may be able to better handle such dynamics.
Numerous resource management schemes have been proposed in order to address
heterogeneously [3], high service demands, fairness, and starvation, as well as transmission
errors due to channel congestion. In spite of this, there are many schemes out there that
prioritize voice services while allocating the remaining bandwidth to non-real-time applications
without any guarantee that the delay will not affect the voice service.
In wireless communication, when multiple resources exchanges between heterogeneous network
environment there is a great chance of distribution of data without the notice of the relevant
agencies. This lost of data sometimes create a big hassle and situation becomes aggravate [4].
In this situation, It is imperative to develop a system that should detect the leaked data
proactively. After detection of the leaked data the prevention measures can be taken for future.
After going through different studies, no specific study is found that can detect the wastage of
REVISTA INNOVACIÓN Y SOFTWARE VOL 4 Nº 2 Septiembre - Febrero 2023 ISSN Nº 2708-0935
54
data during resource transformation mechanism in wireless communication therefore, an
intelligent data detection mechanism has been developed using supervised and unsupervised
testing mechanism. This technique utilized hidden layers of NN and then generates the outcome
which further processed by conducting simulation in Python.
The following are the main contributions of this work.
• Using intelligent supervised and unsupervised testing, the amount of data leakage
would be identified.
• The tested data would be simulated in python to identify the amount of accurate
detected data and the amount of lost data.
• The targeted and actual outcome is analyzed to deduce the impact of proposed method.
Rest of the manuscript is arranged as: Section 2 presents a comprehensive overview of previous
work. Section 3, is enriched with proposed methodology using supervised and unsupervised
testing and Section 4 assesses viability of performance of the proposed method. Finally, Section
5 concludes about findings and future research outlook.
Literature review
It is necessary to advance wireless communication technologies both in terms of scale and
complexity to support emerging applications. Machine learning and predictive regression methods
[5], have been investigated in order to solve wireless resource allocation problems. Supervised
learning involves training neural networks to approximate a function or algorithm so that
computation time is minimized. Additionally, their use requires strategies that protect privacy
while fulfilling the application requirements.
According to Jorge Cort´es [6], privacy preserving data analysis must take into account dynamic
data as well as data exchanged across networks, as well as systems and control perspectives. In
order to protect their data against adversaries with arbitrary side information, they adopted
differential privacy mechanisms that were initially used to analyze large, static datasets.
Under differential privacy constraints, they reviewed how multiple agents can perform signal
estimation, consensus, and distributed optimization tasks. Several factors were ignored in this
study, including how to deal with signals when multiple spectrums exist and how to avoid tangling
them. Further, it is crucial to investigate the appropriate scales for privacy parameters based on
specific application domains as well.
Li Ming [7], work was to identify the current situation of small- and medium-sized organizations'
human resources, using deep learning data. Through the deep learning approach, human
resources can be more productive, and business volumes can be reduced, thereby improving
REVISTA INNOVACIÓN Y SOFTWARE VOL 4 Nº 2 Septiembre - Febrero 2023 ISSN Nº 2708-0935
55
human resource efficiency. His proposed model was improved by implementing a deep neural
network. In an analysis of experimental data, several types of decent gradient processes were
considered as well as a number of neurons in the hidden layer.
Because of the low calculation complexity and fast training speed, the model fails if there are
non-linear relationships between variables, which is common for multiple frequencies.
According to Haoran Sun [8], wireless resource management can be improved by utilizing a
learning-based approach. An unknown non-linear map is treated as a resource allocation
algorithm and approximated using deep neural networks. As long as a DNN of moderate size is
capable of learning accurate and effective non-linear mapping, such a DNN can be used to allocate
resources in almost real time, since the input to get the output only involves a few simple
operations. In order to approximate some of the algorithms of interest in wireless
communications, they developed DNNs to approximate a class of 'learnable algorithms'. Using
numerical simulations, the authors demonstrated that DNNs are superior to conventional
algorithms for estimating two relatively complex algorithms for power allocation in wireless
transmissions. This model only represents a very preliminary step towards understanding the
capability of DNN and how to deal with challenging problems such as beamforming for IC/IMAC
is still unknown. It also could not answer how to further reduce the computational complexity of
DNN?
The application of differential privacy is well documented. However, wireless resource allocation
schemes with differential privacy based on Deep Neural Networks (DNNs) [9], have never been
studied. This study investigates the impact of Differential Privacy DP) [10], on model convergence
and network performance using neural network resource allocation schemes.
These research findings show differentially private schemes can produce high-performance
models, especially when Convolutional Neural Networks (CNNs) [11], are used.
Proposed method using supervised and unsupervised testing
Differential privacy is achieved by adding noise to the data used in a machine learning model in
a way that does not significantly affect the accuracy of the model. This noise ensures that
individual data points cannot be easily identified in the model's output. In neural networks, DP
can be incorporated into the training process by adding noise to the weights or gradients of the
model during training.
There are many functionally interconnected neurons in a neural network that follow a topological
structure [12]. The hierarchical structure of neuron is used to categorize neural network models
into hierarchical and interconnection models.
REVISTA INNOVACIÓN Y SOFTWARE VOL 4 Nº 2 Septiembre - Febrero 2023 ISSN Nº 2708-0935
56
According to a hierarchical model, neurons are classified into different layers in accordance with
their functionality and are interconnected every year. An input, middle, and output layer structure
is used in a hierarchical model, while any two neurons are linked in an interconnection model.
Hierarchical models are widely used because of their easy analysis and good structure.
3.1 Selection Process
Regression problems include prediction of leaked data during resource allocation process. A
regression model is a way of predicting an outcome in machine learning [13]. A regression model
may be linear, elastic, neural, polynomial, or ridge. There is a wide range of models which use
linear variables, among the most common of which is linear regression. The relationship between
the dependent variable and the independent variable is necessary in a linear regression model.
This makes linear regression applicable only to problems whose solutions are linearly separable.
The model estimates linear relationships between variables because it's simple to calculate and
fast to train. Linear regression models are sensitive to outliers, which is a disadvantage.
3.2 Identification of hidden layers
There are several mathematical iterations to calculate amount of actual hidden layers [14]. Two
of the most commonly implemented iterations are represented by equations (1) and (2).
In these equations, the numbers of input and output layer nodes are represented by ni
and no as shown in figure 1. During training, a slight expansion in space is observed, i.e., in the
range and the deep neural network should train continuously based on equation (2). 16, 17, and
1 neurons in total were assessed in the input, hidden, and output layers, respectively. The neuron
activation function represents sigmoid functions, and the error function is seen to be the quadratic
mean square value.
𝑁ℎ = 𝑛𝑖 ×𝑛𝑜
(1)
𝑁ℎ = 𝑛𝑖 ×𝑛𝑜 +K
(2)
REVISTA INNOVACIÓN Y SOFTWARE VOL 4 Nº 2 Septiembre - Febrero 2023 ISSN Nº 2708-0935
57
In supervised testing, n1 as input layer is applied it generates the sublayers as x1, x2.....x5.
These layers further undergo for analyze the actual amount of the hidden layers. Iteration by
iteration, the hidden layers are identified. These hidden layers actually carries the data that has
been separated during resource sharing process. These hidden layers further match the carrying
data with sample data and if it matches, the layers change to l, m, and n in this case. The l, m,
and n layers check the authenticity of the data and save off the ambient data in the form of the
noises. Consequently, final data is transferred to the output layer no.
In next stage the unsupervised testing is carried out. In unsupervised testing, the data is not
labeled and it appears as a mixture of heterogeneous raw table. Figure 2, illustrate the case
where at stage one different data has been mixed and ready to transfer to another network. In
stage 2, from the mixture of data, some data is labeled as a1, a2, a3 and a4 while other data is
transferred. The labeled data is screen out from the hidden layers and the statistics of accuracy
and the lost, from both supervised and unsupervised technique is placed in table 1 and table 2
respectively. The experiment is performed repeatedly to get all detected data. Based on 750
supervised samples and 250 unsupervised samples, Table 1 illustrates the accuracy values.
Observe the final average value of accuracy outcomes after training each hidden layer neuron 25
times consecutively. The execution is terminated once it has run for 500 iterations.
Figure 1. Data leakage detection through Supervised
testing mechanism
REVISTA INNOVACIÓN Y SOFTWARE VOL 4 Nº 2 Septiembre - Febrero 2023 ISSN Nº 2708-0935
58
As the number of neurons in the hidden layer increases, both supervised and unsupervised
accuracy increase. As shown in table 2, both supervised and unsupervised neural networks are
capable of predicting lost data. There is a reduction in loss values with an increase in the number
of hidden neurons.
Table 1. Accuracy of leaked data for supervised and unsupervised testing
Hidden
layers
Number of
Iterations
Accuracy for supervised testing
Accuracy for unsupervised
testing
1
30
86.76%
80.67%
2
60
87.62%
80.14%
3
90
88.54%
81.45%
4
120
89.43%
82.65%
5
150
90.67%
83.67%
6
180
90.87%
84.87%
7
210
91.69%
85.95%
8
240
92.78%
86.74%
9
270
92.54%
87.65%
10
300
93.65%
88.62%
11
330
94.75%
89.84%
12
360
94.86%
90.78%
13
390
94.97%
91.62%
14
420
95.46%
92.78%
15
450
95.54%
93.67%
Figure 2. Data leakage detection through Unsupervised
testing mechanism
REVISTA INNOVACIÓN Y SOFTWARE VOL 4 Nº 2 Septiembre - Febrero 2023 ISSN Nº 2708-0935
59
Performance discourse
After conducting experiments in Python [15], the targeted and the actual generated outcomes
are analyzed. The amount of privacy leakage is directly proportional to the batch size and number
of hidden layers. To improve the rate, one must raise the batch size and number of hidden layers.
Figure 3, depicts the curve for targeted and actual outcomes . starting from 20 samples, the
target and actual generated outcome is the same, however, both outcome fluctuate ups and
down. At sample 70, the targeted detection was aimed at 38, and the same was obtained whereas
at samples were reached 120, there was a difference between targeted and actual output. Here
the target was 37 but received nearly 41%. Similarly, in sample 220, the target was 70% but the
actual outcome was reached to 72%. In the end, the actual outcome left behind the targeted
values which was 79% but achieved 92%.
Table 2. Loss of leaked data for supervised and unsupervised testing
Hidden
layers
Number of
Iterations
Loss supervised testing
Loss for unsupervised testing
1
30
0.0099
0.01017
2
60
0.0098
0.01016
3
90
0.0097
0.01015
4
120
0.0096
0.01014
5
150
0.0095
0.01013
6
180
0.0094
0.01012
7
210
0.0093
0.01011
8
240
0.0092
0.01010
9
270
0.0091
0.01009
10
300
0.0090
0.01008
11
330
0.0089
0.01007
12
360
0.0088
0.01006
13
390
0.0087
0.01005
14
420
0.0086
0.01004
15
450
0.0085
0.01003
16
500
0.0084
0.01002
REVISTA INNOVACIÓN Y SOFTWARE VOL 4 Nº 2 Septiembre - Febrero 2023 ISSN Nº 2708-0935
60
Although the non-normalized dataset [16], may introduce some imperfections, these minute
errors are minimized and do not significantly affect the effectiveness of the proposed prediction
model. In addition, the proposed method achieved higher accuracy and prediction efficiency as
well as a faster convergence rate. Consequently, the prediction model is very accurate. A
supervised accuracy of 96.03% was achieved using the proposed method, while an unsupervised
accuracy of 94.53% was achieved.
Conclusion
The outcome of this study after analyzing the targeted and actual obtained data showed that
using supervised and unsupervised technique to detected the amount of leaked data during data
transfer in wireless network is a hallmark of shrewdness. Models proposed in this study have been
shown to perform better in experiments. The proposed method achieved a supervised accuracy
of 96.03%, while unsupervised accuracy is remained at 94.53%. Data leakage can be identified
using this method with considerable ease.
Contributor Roles
Figure 3. Target versus Actual leaked data
analysis
REVISTA INNOVACIÓN Y SOFTWARE VOL 4 Nº 2 Septiembre - Febrero 2023 ISSN Nº 2708-0935
61
Shahzad Ashraf: Conceptualización, Investigación, Visualización, Metodología, Software,
Validación, Redacción - borrador original, Escritura, revisión y edición. Zeeshan Aslam: Análisis
formal.
Referencias
[1]
R. Zeeshan, and A. Muhammad, “Adopting proactive results by developing the
Shrewd model of pandemic COVID-19,” Arch. Community Med. Public Health, vol.
8, no. 2, pp. 062–067, Apr. 2022, doi: 10.17352/2455-5479.000175.
[2]
Z. Rasheed, S. Ashraf, N. A. Ibupoto, P. K. Butt, and E. H. Sadiq, “SDS: Scrumptious
Dataflow Strategy for IoT Devices in Heterogeneous Network Environment,” Smart
Cities, vol. 5, no. 3, pp. 1115–1128, Sep. 2022, doi: 10.3390/smartcities5030056.
[3]
S. Ashraf, “Avoiding Vulnerabilities and Attacks with a Proactive Strategy for Web
Applications,” vol. 3, no. 2, p. 9.
[4]
Z. A. Arfeen, T. Ahmed, S. Ashraf, and S. Saleem, “Succulent link selection strategy
for underwater sensor network,” Int. J. Comput. Sci. Math., vol. 15, no. 3, p. 224,
2022, doi: 10.1504/IJCSM.2022.10049407.
[5]
M. Gao, Z. Chen, H. Naeem, and T. Ahmed, “CED-OR Based Opportunistic Routing
Mechanism for Underwater Wireless Sensor Networks,” Wirel. Pers. Commun., vol.
125, no. 1, pp. 487–511, Jul. 2022, doi: 10.1007/s11277-022-09561-w.
[6]
J. Cortés, G. E. Dullerud, S. Han, J. Le Ny, S. Mitra, and G. J. Pappas, “Differential
privacy in control and network systems,” in 2016 IEEE 55th Conference on Decision
and Control (CDC), Dec. 2016, pp. 4252–4272. doi: 10.1109/CDC.2016.7798915.
[7]
L. Ming, “A Deep Learning-Based Framework for Human Resource
Recommendation,” Wirel. Commun. Mob. Comput., vol. 2022, pp. 1–12, Jul. 2022,
doi: 10.1155/2022/2377143.
[8]
H. Sun, X. Chen, Q. Shi, M. Hong, X. Fu, and N. D. Sidiropoulos, “Learning to
optimize: Training deep neural networks for wireless resource management,” in
2017 IEEE 18th International Workshop on Signal Processing Advances in Wireless
REVISTA INNOVACIÓN Y SOFTWARE VOL 4 Nº 2 Septiembre - Febrero 2023 ISSN Nº 2708-0935
62
Communications (SPAWC), Sapporo: IEEE, Jul. 2017, pp. 1–6. doi:
10.1109/SPAWC.2017.8227766.
[9]
A. Shahzad, “Towards Shrewd Object Visualization Mechanism,” Trends Comput.
Sci. Inf. Technol., pp. 097–102, Nov. 2020, doi: 10.17352/tcsit.000030.
[10]
D. Muhammad, M. A. Khan, and T. Ahmed, “Fuzzy based efficient Cosmetology
Paradigm,” vol. 8, pp. 513–520, doi: 10.14741/ijmcr/v.8.4.3.
[11]
S. Saleem, and S. Afnan, “FTMCP: Fuzzy based Test Metrics for Cosmetology
Paradigm,” Adv. Comput. Intell. Int. J. ACII, vol. 4, no. 7, pp. 1–13, 2020, doi:
10.5121/acii.2020.7401.
[12]
S. Saleem, S. Ashraf, and M. K. Basit, “CMBA - A Candid Multi-Purpose Biometric
Approach,” ICTACT J. Image Video Process., vol. 11, no. 01, p. 6, 2020, doi:
10.21917/ijivp.2020.0317.
[13]
T. Ahmed, “Sagacious Intrusion Detection Strategy in Sensor Network,” in 2020
International Conference on UK-China Emerging Technologies (UCET), Glasgow,
United Kingdom: IEEE, Aug. 2020, pp. 1–4. doi:
10.1109/UCET51115.2020.9205412.
[14]
S. Saleem, T. Ahmed, Z. Aslam, and M. Shuaeeb, “Iris and Foot based Sustainable
Biometric Identification Approach,” in 2020 International Conference on Software,
Telecommunications and Computer Networks (SoftCOM), Split, Hvar, Croatia: IEEE,
Sep. 2020, pp. 1–6. doi: 10.23919/SoftCOM50211.2020.9238333.
[15]
T. Ahmed, Z. Aslam, D. Muhammad, A. Yahya, and M. Shuaeeb, “Depuration based
Efficient Coverage Mechanism for Wireless Sensor Network,” J. Electr. Comput. Eng.
Innov. JECEI, vol. 8, no. 2, pp. 145–160, 2020, doi:
10.22061/jecei.2020.6874.344.
[16]
A. Yahya et al., “Underwater routing protocols: Analysis of link selection
challenges,” AIMS Electron. Electr. Eng., vol. 4, no. 3, pp. 234–248, 2020, doi:
10.3934/ElectrEng.2020.3.234.