Published Paper Details:

With the rapid expansion of digital payments and e-commerce platforms, online financial transactions have become deeply integrated into everyday life. This growth, however, has been accompanied by a significant rise in fraudulent activities, including identity theft, card-not-present transactions, account takeovers, and phishing-driven attacks. Conventional rule-based fraud detection systems struggle to adapt to evolving fraud patterns and often produce high false-positive rates. To address these challenges, this research proposes TransGuard-AI, an Artificial Intelligence (AI)-driven fraud detection framework that employs supervised machine learning techniques to identify anomalous transaction behavior. The proposed system integrates Logistic Regression, Decision Tree, and Random Forest models, enabling comparative evaluation and ensemble-based insights. A structured feature engineering pipeline is designed to extract transactional attributes such as amount frequency, spending velocity, merchant category patterns, and geolocation deviation scores. Experiments were conducted using a benchmarked credit card fraud detection dataset containing anonymized real-world transaction records with highly imbalanced class proportions. To mitigate class imbalance, Random Under-Sampling (RUS) and Synthetic Minority Oversampling Technique (SMOTE) were applied. Model performance was evaluated using accuracy, precision, recall, F1-score, ROC-AUC, and confusion matrix analysis. Results show that Random Forest outperformed the baseline models, achieving high detection accuracy and improved recall for the minority (fraud) class, while effectively reducing false alarms. Logistic Regression demonstrated faster inference suitability for real-time processing, whereas Decision Tree offered interpretability and rule extraction capabilities. Overall, TransGuard-AI presents a robust and scalable approach for real-time fraud detection in financial systems. The integration of machine learning algorithms, advanced preprocessing, and anomaly-centric feature engineering significantly enhances detection capability, making the system suitable for deployment in modern transaction monitoring infrastructures.