Web-Based Real-Time Phishing Detection System Integrated with Machine Learning

Abstract

Phishing attack prevention that relies on URL blacklist/whitelist analysis combined with machine learning techniques has proven highly effective. However, limitations still exist in the practical implementation of machine learning modules under real-world environments, particularly due to the lack of real-time deployment. Therefore, this research developed a web-based phishing detection system integrated with a machine learning module to analyze, verify, and determine whether to allow or deny access to websites. The system is deployed on a proxy server developed in Python and analyzes (1) domain features, (2) URL features, and (3) path and filename characteristics. It is designed with a user interface (UI) that supports model training and evaluation as needed and accommodates multiple algorithms, including Random Forest (RF), Support Vector Machine (SVM), Multilayer Perceptron (MLP), Deep Neural Network (DNN), and Extreme Gradient Boosting (XGBoost). The evaluation results revealed that the Random Forest (RF) algorithm achieved the highest performance, with an accuracy of 96.70% and an F1-score of 96.68%. Moreover, a user satisfaction survey of 10 participants indicated a high level of satisfaction mean = 4.73, particularly in system reliability and stability, 4.91, while the UI capability for model uploading and training received a relatively lower score of 4.36.