AN OPTIMIZED LSTM MODEL FOR PHISHING WEBSITE DETECTION USING PARTICLE SWARM OPTIMIZATION AND HYPEROPT TECHNIQUES

Abstract

Phishing remains a prevalent cybersecurity threat that exploits human trust to steal sensitive information. Traditional detection methods, such as blacklisting and rule-based approaches, often fail to adapt to the rapidly evolving nature of phishing websites. In contrast, machine learning and artificial intelligence offer powerful solutions by identifying phishing patterns based on URL structures and website behavior. While hyperparamter tuning is a crucial step in machine learning, its impact on phishing detection models remain under-examined, highlighting a need for more research in this area. This study addresses this gap by developing an LSTM-based phishing detection model and optimizing it using two hyperparameter tuning techniques: Particle Swarm Optimization (PSO) and HyperOpt. The results demonstrate that HyperOpt outperforms PSO, achieving an accuracy of 93.12% compared to 92.00% with PSO. This superiority is attributed to Bayesian optimization and the Tree-structured Parzen Estimator (TPE), which enable more efficient hyperparameter selection. The findings emphasize the importance of hyperparameter tuning in improving phishing detection accuracy and enhancing cybersecurity defenses against evolving threats.