Optimizing URL Phishing Detection: A Manifold Learning Approach with an Efficient Neural Network Focused on Reducing Computational Cost

The growing number of phishing attacks is one of the top concerns of cybersecurity researchers. Cryptographic methods are not reliable in stopping phishing attacks because they manipulate the users into thinking it is safe to access a web link. This results in compromise of sensitive information. Recent research has shown how machine learning methods are capable of detecting malicious websites. However, machine learning models can be computationally heavy to produce good accuracy. This paper focuses on reducing the computational cost while maintaining an effective prediction model. The experiment is conducted on a dataset of size (30, 10000), consisting of phishing and benign website links. The Isomap algorithm is implemented to reduce the input dimension while preserving correlation. Subsequently, a neural network is employed, utilizing simple activation and optimization functions to ensure efficient and rapid prediction. The model achieves an accuracy of 87% under 30 training epochs.