A Phishing Detection System by Application of Multi-Classifiers Using E-Voting Method

Phishing remains a major global threat, exploiting human weaknesses to access sensitive data. Existing detection methods often struggle with high false positives and fail to adapt to evolving phishing tactics. This study proposes a phishing detection system that combines Support Vector Machine (SVM), Feed Forward Neural Network (FFNN), and Extreme Learning Machine (ELM) using a weighted voting approach. A cleaned and normalized dataset of URLs was used, with dimensionality reduction via PCA. The models were evaluated using metrics like accuracy, sensitivity, and detection time in MATLAB (R2023a). Results show that SVM achieves the best performance, with the lowest false positive rate (1.79%), highest precision (97.98%), and accuracy (97.75%). FFNN offers balanced performance, while ELM is the fastest but less accurate. The weighted voting mechanism consistently identifies phishing as the dominant class, enhancing detection accuracy. Overall, combining the three models improves robustness, with SVM emerging as the most effective classifier.