OEC[sbnd]Net: Optimal feature selection-based email classification network using unsupervised learning with deep CNN model

Abstract

The escalating prevalence of email communication in global interactions underscores the critical need for robust security measures. However, the existing methods to safeguard email communication, including spam, phishing, and ham emails, have exhibited suboptimal classification performance and inadequate security. So, this work introduces the Optimized Email Classification Network (OECNet) to categorize emails and mitigate security risks effectively. Leveraging a hybrid dataset comprising spam, phishing, and ham email information, we perform meticulous preprocessing to ensure data uniformity. Our approach incorporates principal component analysis (PCA) to extract email-specific pattern properties, enhancing feature discernment through eigenvalue analysis. Inspired by natural swarm behaviours, we employ Particle Swarm Optimization (PSO) to select highly correlated features from PCA outcomes, optimizing the model's efficacy. The OECNet uses PSO-optimised features to integrate a Deep Learning Convolutional Neural Network (DLCNN) model for classification. This novel methodology surpasses traditional methods, yielding an accuracy of 98.43 %, precision of 97.78 %, recall of 96.41 %, and an impressive F1-score of 97.07 %. The proposed OECNet advances email classification technology and establishes a robust security framework against malicious intrusions in global communication. The hybrid dataset incorporates UCI, CSDMC, and SpamAssassin information, ensuring comprehensive coverage and realistic scenarios for effective model training and evaluation.