THE DETECTION OF FRAUDULENT TRANSACTIONS USING MACHINE LEARNING

Abstract

Credit cards are massively used nowadays for internet transactions performed at any moment, given that they have offered facilitation both in usage and time. With the growing usage of credit cards, there has also been an increase in their misuse capacity. Credit card deceits cause considerable financial loss not only for their owners but also for the financial companies. The main objective of this research study is the identification of the fraudulence cases which may include the access of the public data, handling groups of largely destabilized data and the adaption to the developing deception models. The corresponding literature poses many approaches based on Machine Learning for the detection of credit cards, some of which are: Extreme Learning Method, Decision Tree, 0Random Forest, Support Vector Machine, Logistic Regression and XG Boost. However, due to an insufficient accuracy, there is still some need to apply deeper algorithms to reduce the loss from fraudulence. For this aim, the main focus of this research wok has been the application of “Deep Learning” algorithms. A comparing analysis between the two algorithms “Machine Learning” and “Deep Learning” was conducted in order to retrieve efficient results. Also, a Machine Learning algorithm was applied on the group of data, which improved significantly the accuracy of detecting fraudulence. Moreover, I applied three architectures based on a convolutional neural network to ameliorate even further the performance of fraud detection. A complete empirical analysis was performed by experimenting with different configurations of the hidden layers by changing the number of training epochs and using the latest models. The findings from this research demonstrate enhanced results, specifically in terms of accuracy and precision. The suggested model outperforms the most recent Machine Learning and Deep Learning algorithms designed for addressing credit card fraud detection issues. In addition, I conducted experiments to balance the data and implemented Deep Learning algorithms to reduce the occurrence of biased negative results. These proposed methods can be efficiently employed to identify instances of credit card fraud in real-world scenarios.