Intelligent Image Spam Analysis Using CNN-Based Visual Semantics and Web Log Mining for Next-Generation Spam Filtering
Year: 2024
Type: article
Abstract: Image-based spam presents a persistent challenge to traditional text-oriented filtering systems due to its use of visual obfuscation techniques. This paper proposes an intelligent image spam detection framework that combines convolutional neural network (CNN)-based visual semantic analysis with web log mining to improve detection accuracy and robustness. The framework processes image content and associated transmission logs in parallel, extracting high-level visual features using a deep CNN backbone and behavioral patterns from web and sender logs. These heterogeneous features are integrated through a late-fusion classification strategy to produce reliable spam predictions. Experimental evaluation on publicly available image spam datasets, augmented with adversarial obfuscations, demonstrates that the proposed multimodal approach consistently outperforms visual-only and behavior-only baseline models across standard performance metrics, including accuracy, F1-score, and ROC–AUC. Ablation studies highlight the significant contribution of web log features in enhancing robustness, while explainability analysis using SHAP and Grad-CAM provides transparent insights into model decisions. The results confirm that integrating visual semantics with behavioral context offers an effective and scalable solution for next-generation image spam filtering systems.
Author: Khedkar Vaishali Shankar
Language: English
Topic: Spam and Phishing Detection
Subfield: Information Systems
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W7217526348
