Enhancing Phishing Website Detection with Feature Credibility Score and Weighted Threshold using Neural Networks

Phishing attacks continue to pose a significant threat to individuals and the broader online ecosystem, exploiting sophisticated tactics to deceive users into revealing sensitive information. Leveraging the capabilities of neural networks, which excel at learning from extensive datasets, has shown promised results in detecting these malicious phishing websites. However, during the training of neural network models, they often encounter challenges related to the underrepresentation of certain features, which can limit their overall effectiveness. We introduce a Feature Credibility Score (FCS) and a Weighted Threshold to facilitate the selection of the most appropriate features. Feature Credibility Score (FCS) highlights significant features based on their correlation with the target variable, enhancing the model's interpretability. Meanwhile, Weighted Threshold adapts feature selection according to the dataset's nuances, offering a tailored approach for optimal performance. Together, these advancements address the challenges of overfitting, improve feature selection, and offer a customizable approach to optimizing model performance, ultimately strengthening the cybersecurity defence against phishing attacks.