A Robust and Efficient Machine Learning Approach for Identifying Fraud in Credit Card Transaction

The rise in credit card fraud in this age of technological advancements calls for the creation of trustworthy and efficient detection systems. This work investigates how to enhance credit card fraud identification using ML models, with an emphasis on ensemble approaches. A thorough analysis of the literature to date has shown that fraud detection algorithms now in use have several drawbacks, including data imbalance, idea drift, false positives and negatives, limited generalizability, K-fold validation, robustness, and challenges with real-time processing. This study investigates the use of pre-processing and sampling strategies for machine learning models-based credit card fraud detection. The study employs several strategies, such as missing data handling, rejecting outliers, feature selection, under sampling, and oversampling, to balance and improve the quality of the dataset. Later, five prominent ML models, namely, logistic regression, support vector machine, decision tree, extreme gradient boosting, and k-nearest neighboring, were applied to under-and over-sampled data separately to classify credit card fraud detection. The project was implemented in Google Collaboratory with the support of the graphics processing unit (GPU). These sample processes were thoroughly compared, and 5-fold cross-validation was used to assess the model’s performance using the accuracy metrics and F1 score. The results demonstrate that models trained on oversampled data outperform those trained on under sampled data, with XGBoost achieving the highest accuracy (98.37%) and F1 score (97.95%) among the models. The robustness and consistency of the models were confirmed using multiple data K-fold validation, highlighting the effectiveness of the employed techniques. Additionally, the models’ learning curves were used to assess the models’ bias. This study highlights the value of pre-processing and sampling in the development of reliable and strong credit card fraud detection systems by introducing a unique data set with 22 real-time characteristics.