Improving Fraud Detection Accuracy in Mobile Payment Systems

The aim of this study is to improve accuracy of fraud detection in mobile payment systems through comparison of K-Nearest Neighbors (KNN) and Artificial Neural Networks (ANN).Fraudulent transactions pose the biggest danger to financial security, and there is a need for high-accuracy machine learning models with high performance in terms of detecting such transactions. A Kaggle dataset of transaction records that were fraudulent or genuine was utilized. The research employs the two-group experiment design: Group 1 uses the ANN model, which utilizes multi-layer perceptron (MLP) structure to identify intricate patterns of fraud, Group 2 uses the KNN algorithm combined with principal component analysis (PCA) for reducing dimensionality.The models' performance was measured for accuracy, precision, recall, and F1-score. The validity was conducted through a statistical test on G Power with 80% value, p-value 0.05, and the confidence level is 95%.ANN model outperformed KNN with higher accuracy (93% vs. 89%), lower variability (SD = 0.06 vs. 0.09) and better fraud detection ability, which is reflected in its higher F1-score and precision trend.The findings confirm that ANN performs better than KNN in detecting fraud since it possesses the ability of learning complex fraud patterns. Future research must focus on optimization of ANN for real-time detection and studies on hybrid models to provide further security.