Adaptive Credit Card Fraud Detection Using Machine Learning and Deep Reinforcement Learning

Credit card fraud detection is a challenge in the financial sector, where the rarity of fraudulent transactions makes accurate classification particularly difficult. This study presents a comprehensive approach that integrates data preprocessing, resampling techniques, traditional machine learning models, anomaly detection methods, and deep reinforcement learning for effective fraud detection. Initially, extensive exploratory data analysis (EDA) was conducted, followed by handling missing values and applying Synthetic Minority Over-sampling Technique (SMOTE) to address class imbalance. A variety of supervised models, including Logistic Regression, Random Forest, XGBoost, and Multi-Layer Perceptron (MLP), as well as unsupervised anomaly detection methods like Isolation Forest and Local Outlier Factor, were evaluated. Subsequently, a Deep Q-Learning Network (DQN) was implemented to model fraud detection as a sequential decision-making problem, allowing the system to dynamically learn fraud patterns. The experimental results demonstrate that DQN achieved high precision, recall, and F1- score, outperforming several traditional classifiers. This study highlights the importance of combining classical and modern learning paradigms to enhance information assurance in credit card transaction systems. The code supports reproducibility and future research.