Deteksi Phishing Website Menggunakan Algoritma Random Forest dengan Hyperparameter Tuning GridSearchCV

Abstract

Phishing websites are one of the cybersecurity threats that deceive users into disclosing sensitive information such as usernames, passwords, and financial data. These attacks are commonly carried out by creating fake web pages that resemble legitimate websites, making them difficult for ordinary users to identify. Therefore, URL-based phishing detection is important in improving user security when accessing websites. This study aims to improve the accuracy of phishing URL classification by applying hyperparameter tuning to the Random Forest algorithm as a popular machine learning method. The dataset used consists of 10,000 URLs, comprising 5,000 phishing URLs and 5,000 legitimate URLs, which were then extracted into 16 relevant numerical technical features. Four model configurations were evaluated, consisting of one default model and three tuned models using GridSearchCV. The experimental results show that the Grid 2 configuration achieved the best performance with an accuracy of 82.65%, precision of 82.70%, recall of 82.65%, F1-score of 82.62%, ROC-AUC of 0.9068, and PR-AUC of 0.8976. In comparison, the default model only achieved an accuracy of 82.40% and an F1-score of 82.38%, while Grid 1 and Grid 3 produced slightly lower performance. These findings indicate that hyperparameter tuning can provide a slight performance improvement compared to the default configuration in detecting phishing websites, although the difference is not statistically significant.