Implementasi Model Machine Learning untuk Deteksi Phishing dengan Pendekatan Ekstraksi Fitur yang Dioptimalkan

Abstrak

Phishing is a common form of cybercrime used by digital criminals to steal sensitive information such as passwords, personal data, and financial details through fake websites designed to re-semble legitimate pages. However, conventional detection methods such as blacklists and manual inspection are currently considered ineffective due to their static nature, often failing to recognize new, evolving and increasingly sophisticated attack patterns. To address this issue, this study developed a machine learning-based phishing detection model focused on improving the accura-cy and efficiency of identifying malicious sites. This model applies an optimized feature extrac-tion technique to enable the system to analyze URL characteristic patterns more comprehensively and targeted. The research dataset was taken from the Kaggle platform, which provides a dataset of phishing and benign URLs with a high reputation. The data was then processed through nor-malization, cleaning, and extraction of important features such as URL structure and domain at-tributes. The classification process was carried out using an ensemble learning approach that combines four popular algorithms: Random Forest, Gradient Boosting, Logistic Regression, and AdaBoost through a soft voting mechanism. The evaluation results show that the proposed model has excellent performance with an accuracy of 98.10%, a precision of 97.81%, a recall of 93.90%, an F1-Score of 95.82%, and a ROC-AUC of 98.62%. These findings confirm that the ensemble ap-proach with optimized features has great potential for application in artificial intelligence-based cybersecurity systems capable of adaptive and real-time phishing detection.