ARCHIVES
VOL. 7, ISSUE 2 (2023)
Enhanced predictive data mining algorithm for fraud detection and churn behaviour modelling in telecommunication systems
Authors
Promise Elechi, Iwok Odudu-Abasi Michael
Abstract
An improved data mining predictive model for
fraud detection and churn behavior in a telecommunication network system is
presented in this thesis. The large revenue losses suffered by telecom service
providers as a result of fraud served as the impetus for this study.
Telecommunications fraud detection and prevention includes any strategy or
procedure used to minimize illicit activity meant to hurt telecom service providers.
The losses include anything from the price of persuading a brand-new client to
use the services of the supplier to the price of keeping hold of current
clients. The complicated network architecture that categories and retains
derived patterns in the cloud back-end for analytics was the subject of the
study. To develop an adaptive control strategy for fraud detection,
computational analytic modelling was used in conjunction with probabilistic
models, a Naive Bayesian model, a linear discriminant function, and neural
prediction networks. To take use of call detail records travelling
non-homogeneously through the network, simulation and train-classification
technique were investigated. Modularization, class containers, and data
structures were used to create the Java-SQL Containerization technique. When
describing the communication network as having edge devices coupled to the Base
transceiver (BT) controllers, the global system architecture for Fraud behavior
synthesis was described. The subscriber/sample radial basis neural network
function (SRBNF) was constructed from the network using computational neural
controller architecture. A predictive probability data mining model was created
during the data analysis employing the Non-Homogenous Poisson Process (NHHP
(t)) for prior knowledge. The posterior probability was calculated using the
Naive Bayes (supervised learning) classification algorithm for low- and
high-income subscribers. For a multivariate analysis, a critical threshold
discriminant function (CTDF) Value of 0.00229 was achieved. As a result, a
pooled sample dispersion matrix was created, as well as inverse dispersion
matrices for the neural computational model that was created. The proposed data
mining predictive model provided a Mean Square Error (MSE) for the CTDF of
1.7562 in the SRBNF validation. To find the best algorithm or model that
produces accurate, dependable outcomes consistently, an examination was
conducted. Therefore, three algorithms—Decision Tree (DT), Logistic Regression
(LR), and Enhanced Neural Discriminant Analysis (Proposed)—were examined for
fraud detection. These yielded, respectively, 14.29%, 30.00%, and 55.71%. The
best and most accurate prediction threshold for fraud attrition was thus found
to be provided by the suggested method.
Download
Pages:1-4
How to cite this article:
Promise Elechi, Iwok Odudu-Abasi Michael "Enhanced predictive data mining algorithm for fraud detection and churn behaviour modelling in telecommunication systems". International Journal of Advanced Engineering and Technology, Vol 7, Issue 2, 2023, Pages 1-4
Download Author Certificate
Please enter the email address corresponding to this article submission to download your certificate.