A DEEP LEARNING ENSEMBLE WITH DATA RESAMPLING FOR CREDIT CARD FRAUD DETECTION

Abstract

Credit Cards are so important in today's digital economy. Their use has expanded a lot in the last few years, which has resulted in a rise in credit card fraud. Machine learning (ML) techniques have been used to detect the credit card fraud. But as shopping habits of credit card holders change all the time and there is an imbalance between classes, it has been difficult for ML classifiers to work at their best. To overcome this problem, in this research, the resilient deep learning methodology has been introduced, in which, on the one hand, long short-term memory (LSTM) and gated recurrent units (GRUs) neural networks are used as the initial learners, and on the other hand, multilayer perceptron (MLP) is employed as the meta-learner which is blended within the framework of a stacking ensemble architecture. At the same time, the hybrid synthetic minority oversampling technique and edited nearest neighbour (SMOTE-ENN) method are used to enable making the class distribution in the dataset more even. The experimental results showed that the combination of the proposed deep learning ensemble and the SMOTE-ENN approach achieved a sensitivity of 1.000 and accuracy specificity of 0.997 to overcome other machine learning classifiers and methodologies that are popular and documented in the literature. Next, we discuss slightly more elaborate ensemble methods such as Stacking and Voting Classifier and test them on both the original and SMote-ENN datasets. Also, a flask framework with SQLite integrated in it allows for users to sign up, sign-in, and test things so the project can work better and make it possible for users to interact with the project better.