Hyperparameter Optimization for Malicious URL Detection: Leveraging Optuna and Random Search in Machine Learning and Deep Learning Models

Uniform Resource Locators (URLs) are critical indicators for identifying malicious online activities such as malware distribution, phishing attacks, and website defacement. This research presents a robust approach for detecting these threats using both deep learning (DL) and machine learning (ML) techniques. We emphasize hyperparameter optimization, employing Optuna—a Bayesian optimization framework— and Random Search to systematically enhance model performance. Unlike many prior studies, which often overlook thorough hyperparameter tuning, our approach demonstrates improvements over state of the art methods. Our Bidirectional Encoder Representations from Transformers (BERT) model achieved an accuracy of 98.84%, with an F1 score of 99.02%, while the Light Gradient Boosting Machine (LightGBM) attained an accuracy of 98.46% and an F1 score of 98.45%