A HYBRIDIZED SMOTE-ENN APPROACH ON IMBALANCED DATASET OF FRAUDULENT CREDIT-CARD SCENARIO

Abstract

Businesses and financial activities are now carried out effortlessly thanks to the advancement of information technology. Credit cards make it simple and comfortable to perform company activities remotely. However, this advancement is not without obstacles and compromises, since credit card fraud is expanding at an exponential rate. Thus, in order to address this difficulty using cutting-edge deep learning technology to detect fraud, the dataset in question must be easily available and balanced. However, most of the available datasets are not balanced, thereby potentially affecting the accuracy of the learning models to detect or classify. To this end, this study aims to hybridize the Synthetic Minority Oversampling Technique-Edited Nearest Neighbor (SMOTE-ENN) algorithm to balance the dataset and detect the possibility of fraud. SMOTE is taken into consideration in order to proffer a solution to the imbalanced nature of the dataset, which was acquired from the Kaggle repository based on the insight of the benchmark literature. The ENN, which is the deep neural network, would in turn receive the output from this process. Based on the results, the hybridized technique is promising because the model was able to record an accuracy and F1-score of 99%.