Leveraging pigeon-inspired optimizer with deep learning model on website phishing detection and classification for secure web mining

With the high growth of the Internet, the security of networks has stimulated individuals' attention. It is believed that a safe system atmosphere is an effective source for the fast and complete expansion of the Internet. Phishing is a vital type of cybercrime, which is a mischievous action of tricking consumers into clicking on phishing links, stealing consumer data, and eventually utilizing user information to fake log in with linked accounts to take assets. The models of phishing and the expertise of recognition are always being upgraded. With the progress and applications of machine learning (ML) technology, numerous ML-based solutions for detecting phishing have been developed. Some solutions depend upon the extraction of features by rubrics, while others require trusting third-party services, which can affect variability and lead to time-consuming issues in the forecasting service. Thus, this article develops a novel Pigeon Inspired Optimizer with a Deep Learning Model on Website Phishing Detection and Classification for Secure Web Mining (PIODL-WPDCWM) algorithm. The objective of the PIODL-WPDCWM technique lies in securing web mining activities and defending users from phishing attacks on websites. Primarily, the presented PIODL-WPDCWM technique involves z-score normalization to ensure that input features are standardized to a common scale. For the feature selection procedure, the brown-bear optimization algorithm (BBOA) has been employed to classify the most relevant and informative features from the data. Additionally, the self-attention-based long short-term memory and auto-encoder (S-LSTM-AE) classifier is deployed for the detection and classification of website phishing. Lastly, the pigeon-inspired optimizer (PIO) algorithm can be utilized for the hyperparameter tuning model of the S-LSTM-AE method. To certify the higher performance of the PIODL-WPDCWM technique, a wide range of simulation studies was conducted, and the attained outcomes demonstrated the improvement of the PIODL-WPDCWM technique over other existing models.