Deteksi Email Phishing Menggunakan Machine Learning Dengan Algoritma Random Forest Classifier dan Random Search Cross-Validation

Email is widely used in educational institutions, but its operational convenience is accompanied by phishing threats that can lead to credential theft, data leakage, and other security incidents. This study develops a machine learning-based system for classifying email text into Safe Email and Phishing Email categories using Random Forest optimized with Random Search Cross-Validation (RSCV). The research uses historical email data from SDN Limo 1 Kota Depok. The processing stages include text cleaning, case folding, masking of email addresses and URLs, stopword removal, stemming, outlier handling, and TF-IDF feature extraction. The dataset is divided into 80% training and 20% testing data. Random Forest is evaluated in a standard configuration and after hyperparameter optimization using five-fold cross-validation and ten random iterations. The optimized model achieves 96.99% accuracy, 97% macro precision, 97% macro recall, and 97% macro F1-score, improving accuracy by 2.51 percentage points over the standard model. The model is implemented in a Streamlit web application to support practical preliminary email screening in educational environments. The application also provides prediction probabilities, an explainable analysis of influential terms, testing history, role-based access, and data export for authorized users. The results indicate that hyperparameter optimization can improve classification performance while the integrated application can support faster initial screening before further security verification is performed.