The effectiveness of machine learning in detecting phishing websites

Phishing poses a significant risk in the field of digital security, requiring effective methods for identifying fraudulent websites. This study evaluated the performance of nine machine learning classification models in the context of phishing website detection. Two different input datasets were prepared: the first included the full HTML code, while the second was based on a set of features extracted from that code. The analysis revealed that models trained on the extracted features achieved nearly twice the detection performance compared to those operating on raw HTML code. The use of majority voting further improved classification effectiveness. The study results confirm that proper feature selection and the integration of outputs from multiple models significantly enhance the effectiveness of systems for detecting online threats.