EVALUATING THE PERFORMANCE OF RANDOM FOREST AND LOGISTIC REGRESSION MODELS IN DETECTING BANK TRANSACTION FRAUD: A COMPARATIVE STUDY

Abstract

This study explores developing and evaluating machine learning models for detecting fraudulent bank transactions. By analyzing transaction data, features such as transaction type, amount, balance, and date are extracted and labeled as genuine or fraudulent based on balance consistency and transaction limits. The dataset is split into training and testing sets, and two models—Random Forest and Logistic Regression—are trained using standardized features. The models are evaluated on accuracy, precision, recall, and F1-score metrics. Results indicate that the Random Forest model outperforms Logistic Regression in terms of accuracy due to its ability to handle complex relationships within the data. However, Logistic Regression offers valuable probabilistic insights. Challenges such as data imbalance and feature extraction quality are addressed with techniques like Synthetic Minority Over-sampling Technique (SMOTE) and advanced preprocessing methods. Prediction probabilities are visualized using Matplotlib for better interpretation. Future work includes enhancing feature extraction, expanding the dataset, and exploring more advanced models to further improve performance. This study demonstrates the potential of combining multiple validation techniques and machine learning models with a userfriendly interface to create a robust solution for detecting fraudulent bank transactions, thereby enhancing financial security