Machine Learning in Financial Security: Performance Evaluation of Fraud Detection Models for POS Operators in Nasarawa Town

Abstract

Financial fraud continues to present a growing challenge for digital payment systems, especially for Point-of-Sale (POS) operators navigating the complexities of emerging economies. In Nasarawa Town, Nigeria, the rapid expansion of POS services has inadvertently created more opportunities for fraudulent transactions, highlighting an urgent need for smarter, more adaptive detection strategies. This study examines how effectively five supervised machine learning models, Logistic Regression, Decision Tree, Random Forest, Support Vector Machine (SVM), and Neural Network, identify fraudulent POS transactions. Using anonymized transaction data, we carefully pre-processed the dataset, engineered relevant features, and addressed the persistent challenge of class imbalance before training and validating each model. Performance was assessed through accuracy, precision, recall, F1-score, and Area Under the ROC Curve (AUC). Results showed that Random Forest consistently led across all metrics, achieving 96.4% accuracy and 97.1% AUC, demonstrating its ability to reliably separate legitimate from fraudulent activity while maintaining a practical balance between catching fraud and avoiding false alerts. Neural Network followed as a strong contender, while Logistic Regression and Decision Tree showed more modest results. We conclude that ensemble learning approaches, particularly Random Forest, offer the most dependable and scalable solution for real-time fraud detection among POS operators in Nasarawa Town. These findings contribute locally relevant evidence to financial fraud analytics and offer actionable insights for developing machine learning–powered security frameworks in semi-urban digital payment environments.