Identification of the Best Combination of Oversampling Technique and Machine Learning Algorithm for Credit Card Fraud Detection

In recent times, digitalization has become increasingly popular due to the seamless and convenient use of e-commerce. Online payment and shopping has become the most preferred method due to the time and transport convenience. However, the widespread use of e-commerce has also led to a significant increase in credit card fraud. Fraudsters are using fake identities and various technologies to deceive users and obtain money illegally. As a result, it has become essential to find a solution to prevent these types of fraudulent activities. In this research study, a model is designed to detect fraud in credit card transactions, which provides important features required to identify illegal transactions accurately. Due to the constantly evolving nature of technology, it is becoming increasingly difficult to track the behaviour and patterns of fake transactions. Therefore, a powerful fraud detection system is necessary for accurately detecting and preventing the fraudulent activities. This study uses a variety of machine learning methods on an unbalanced dataset to address the problem of fraudulent use of credit cards. Naïve Bayes, ANN, decision tree, random forest, and logistic regression are applied to the dataset with oversampling techniques such as K-Means SMOTE, SMOTE, SMOTE-NC, ADASYN, and an undersampling technique, NearMiss, is applied to the data. The performance metrics of each algorithm are evaluated using quantitative measurements, including precision, accuracy, F1 score, and recall. The primary objective of this research study is to develop the best machine learning algorithm that can successfully identify fraudulent behaviours when applied to a dataset by utilizing oversampling methodologies.. The outcomes show that the Random Forest (RF) algorithm using the ADASYN approach has achieved the highest F1 score of 0.816, demonstrating its potential in fraudulent activity detection.