Detection of Phishing Websites using Novel Machine Learning Fusion Approach

Abstract: The Phishing websites continue to pose a serious cybersecurity threat by enabling credential theft, financial fraud, and malware distribution through deceptive online platforms. Conventional phishing detection mechanisms based on blacklists and rule-based heuristics are ineffective against newly emerging and zero-day phishing attacks, as adversaries continuously modify URL structures, webpage content, and obfuscation techniques. This paper proposes a novel machine learning fusion approach for phishing website detection that integrates multi-dimensional feature categories, including lexical, host-based, network, content, visual, and behavioural features. Multiple machine learning classifiers, Decision Tree, Random Forest, Gradient Boosting, and Support Vector Machine, are trained on the engineered feature set, and their predictions are combined using a meta-learning-based decision fusion strategy. Feature normalization and dimensionality reduction techniques are employed to reduce redundancy and improve computational efficiency. The proposed framework is evaluated using benchmark phishing datasets and real-time collected URLs. Experimental results demonstrate that the fusion-based approach achieves superior accuracy, precision, recall, F1-score, and ROC-AUC compared to individual classifiers and traditional detection systems. The findings confirm that hybrid feature engineering combined with intelligent model fusion provides a scalable, robust, and effective solution for detecting sophisticated and zero-day phishing websites.