Performance of Spam Detection Models for Complex Data: Integrative and Hybrid Approaches
Cao Luping, Pengfei Guo · 2025
With the rapid development of information technology, spam poses significant challenges, affecting information transmission and security. This paper addresses spam detection through a combined integrated and hybrid model approach, enhancing classification performance using multiple learners. By integrating a probability-corrected Support Vector Machine (SVM) with an optimized lightweight gradient Elevator (Light- GBM), the model's robustness is improved. TF-IDF was employed for text feature extraction and adversarial sample generation. Experimental results indicate that the Light-SVM integrated model achieves 99.39% accuracy and a 97.68%F1 score, while the hybrid model LSmix attains 98.21 % accuracy and a 92.82% F1 score, demonstrating excellent detection effectiveness, stability, and capability in handling complex data and adversarial samples.