Comparative Machine Learning Classification for QRIS Quishing Detection Based on URL Features

English

Abstract

The increasing adoption of the Quick Response Code Indonesian Standard (QRIS) as a digital payment method has been accompanied by the emergence of quishing, a phishing attack that exploits malicious QR codes to redirect users to fraudulent websites. This study aims to compare the performance of three machine learning classification algorithms Random Forest, Decision Tree, and Naïve Bayes—for detecting phishing URLs in a simulated QRIS quishing environment using URL-based features. The experiments were conducted using a publicly available phishing URL dataset representing simulated URLs that may be encountered after scanning malicious QR codes. Model performance was evaluated using accuracy, precision, recall, and F1-score. The experimental results show that Random Forest achieved the highest accuracy of 96.94%, outperforming Decision Tree 95.32% and Naïve Bayes 65.49%. The superior performance of Random Forest is attributed to its ensemble learning mechanism, which combines multiple decision trees to reduce overfitting, improve robustness, and provide more stable classification performance across diverse URL characteristics. This study contributes a comparative benchmark of machine learning algorithms for URL-based quishing detection and demonstrates that Random Forest is the most effective approach for supporting early phishing detection in QRIS payment systems.