A comprehensive dual-layer architecture for phishing and spam email detection

Abstract

The widespread use of email communication for sharing personal, professional, and financial data has rendered it vulnerable to cyber-attacks. Detecting untrustworthy emails with minimal errors is crucial to counter these threats. This research paper focuses on classifying spam and phishing emails, commonly employed to steal confidential information by impersonating legitimate sources. The scale of such attacks is alarmingly high, causing substantial financial losses across sectors like banking, healthcare, technology, and other businesses. This research aims to classify both spam and phishing emails, addressing the limitations of existing studies that only focus on one type and consider either the email body or content for feature selection. This research incorporates features from both the email body and content during model training. The authors propose a novel approach that ensures highly accurate classification while effectively handling the common issue of data imbalance in email phishing and spam classification. The approach utilizes a dual-layer architecture, with each layer containing a trained or pre-trained model that classifies data instances into their respective classes. Layer 1 classifies the phishing class, while Layer 2 classifies the spam class. Building upon the proven effectiveness of deep learning techniques for text classification and analysis, the proposed architecture employs models like Artificial Neural Networks (ANN), Recurrent Neural Networks (RNN), and Convolutional Neural Networks (CNN). Experimental evaluation demonstrates the approach’s remarkable accuracy, recall, precision, and F1-score, achieving 99.51%, 99.68%, 99.5%, and 99.52%, respectively. This signifies its high efficacy in detecting and classifying malicious emails with minimal errors, thus holding great promise in enhancing system security against cyber-attacks in email communication.