Effective Credit Card Fraud Detection Using Data Mining Techniques

Businesses and consumers around the world face financial and security problems related to credit card fraud. Fraudulent activities are becoming more sophisticated, and therefore the need for effective and/or efficient fraud detection systems has become essential. This study focuses on how machine learning techniques can be applied to detect credit card fraud specifically, and how to overcome challenges like class imbalance, high dimensionality and complexity of real-world data sets. The IEEE-CIS Fraud Detection dataset, a publicly available and highly complex dataset, was utilized to evaluate the performance of various machine learning models. This study compares five machine learning models which are Logistic Regression, Random Forest, XGBoost, LightGBM, and Deep Neural Networks (DNN), to establish a performance baseline using the full dataset with the k-fold stratified cross validation method. Feature engineering was subsequently performed on the best-performing model (LightGBM), utilizing gain-based importance and cumulative feature importance to identify and retain the most relevant features. The reduced dataset was used to retrain the model, and its performance was evaluated against the full dataset to assess the effectiveness of the feature engineering process. An important finding is that feature engineering helped to reduce dataset dimensionality and improve model predictive performance, especially for fraudulent transaction detection. Consequently, the results showcase ensemble methods and advanced feature selection techniques as a possibility for constructing robust fraud detection systems. This research adds to the literature of machine learning applications in the area of fraud detection and it advances our understanding of how to obtain a balance between computational efficiency, interpretability, and accuracy. This study addressed to limitations of the traditional approaches and used state of the art machine learning methodologies in order to provide practical and theoretical contributions to the fight against credit card fraud and for future research and to real world implementations.