Analisis Komparatif Decision Tree, Random Forest, dan Naive Bayes untuk Deteksi Website Phishing Menggunakan K-Fold Cross Validation dan Evaluasi Multi-Metrik

Nasir Usman (STMIK Profesional Makassar)
Muhammad Faisal (Department of Informatics, Universitas Muhammadiyah Makassar, Indonesia)
Alvina Felicia Watratan (Department of Computer Science, STMIK Profesional Makassar, Makassar, Indonesia)
Emil Agusalim Habi Talib (Department of Informatics, Universitas Muhammadiyah Makassar, Indonesia)
Nurahmad (Department of Computer Systems, Universitas Handayani, Makassar, Indonesia)

ROUTERS: Jurnal Sistem dan Teknologi Informasi
Vol. 4 No. 2, Juli 2026 (In Progress)

Phishing websites remain difficult to detect because attackers can create new domains faster than blacklist systems can update. This study compares Decision Tree, Random Forest, and Naive Bayes for phishing website classification using 11,055 records with 30 technical features. The models were evaluated with Stratified 10-Fold Cross-Validation and five metrics: accuracy, precision, recall, F1-score, and ROC-AUC. Random Forest produced the best and most stable performance, with 97.23% accuracy, 97.70% precision, 96.02% recall, 96.85% F1-score, and 99.58% ROC-AUC. Decision Tree also performed strongly, while Naive Bayes showed very high recall but many false positives. Feature importance analysis identified SSLfinal_State and URL_of_Anchor as the most influential predictors, supporting lightweight technical-feature screening for phishing detection.