International Research Journal of Education and Technology

The increased digitization of credit card transactions has led to significant growth in fraudulent activities within the banking sector, creating a pressing need for advanced fraud detection mechanisms.Conventional rule-based methods have become insufficient in handling sophisticated and changing fraud schemes as these methods are static and not dynamic. This project will create a sophisticated fraud detection system focused on credit card transactions, utilizing machine learning algorithms for handling high-volume, real-time transaction data. The proposed system will integrate the use of machine learning algorithms such as Random Forest, XGBoost, and Decision Trees in a robust, fault-tolerant framework. The architecture dynamically switches between algorithms to ensure optimal performance, enhance resilience, and improve accuracy in fraud detection. To address the class imbalance issue that is generally a characteristic of fraud detection data, where legitimate transactions significantly outpace fraudulent ones, this system uses SMOTE to synthetically create samples of fraudulent transactions, thereby improving its ability to detect anomalies. The tasks involved in data preprocessing of this system include missing value imputation, duplicate removal, and data normalization, making the input data clean and suitable for training high-performance machine learning models. Furthermore, feature selection is also used to narrow down the database by picking the most salient transaction attributes like amount, location, and frequency that distinguish between legitimate credit card transactions and fraudulent credit card transactions.

Keywords: Credit card fraud detection, Machine learning, XGBoost, Decision Tree, Random Forest, SMOTE.