Intelligent Deep Ensemble Learning Model for Advanced Phishing Attack Detection and Cybercrime Forensics

Abstract

One of the most common types of cybercrime is phishing, which uses social engineering and technological deceit to obtain sensitive personal and financial data through email spoofing, malicious websites, and destructive software installations. Phishing attack detection is an important field of study in cybercrime forensics because attackers use a variety of communication methods, such as emails, URLs, messaging apps, and phone calls. In order to overcome this obstacle, this work builds a new model that combines group convolution with a symmetric structure rather than using a conventional CNN. It then uses the SMOTE to control the data's class imbalance. Incorporating snapshot ensemble improves the model's generalisability without drastically raising computational costs, while cyclic cosine annealing learning rates further boost the training process. With a classification accuracy of 99.12%, the suggested method for detecting phishing attacks outperforms four existing ensemble methods, according to the results. Using ensemble learning techniques in conjunction with group convolution greatly improves accuracy and adaptability, as seen below. Finally, the suggested approach provides a solid defence against the increasing danger of phishing attempts by providing a dependable, effective, and extensible cybercrime forensics solution.

https://doi.org/10.65470/james.11