INVESTIGATING EVASIVE TECHNIQUES IN SMS SPAM FILTERING

Abstract

The rapid growth of mobile communication has led to a significant increase in SMS spam, posing serious security and privacy risks to users worldwide. Modern spammers continuously employ evasive techniques such as word obfuscation, character substitution, spacing manipulation, and synonym replacement to bypass conventional spam detection systems, making accurate classification increasingly challenging. This paper presents an intelligent SMS spam filtering framework that investigates the effectiveness of multiple machine learning models in detecting spam messages under adversarial conditions. The proposed framework combines comprehensive text preprocessing, Term Frequency–Inverse Document Frequency (TF-IDF) feature extraction, and comparative evaluation of traditional and deep learning classifiers, with particular emphasis on the Long Short-Term Memory (LSTM) network due to its ability to capture contextual and sequential information from textual data. The system is trained and evaluated using a large, balanced SMS dataset containing both legitimate and spam messages collected from diverse sources. Standard performance metrics, including accuracy, precision, recall, F1-score, and confusion matrix analysis, are used to assess the robustness of each model against evolving spam patterns. Experimental results indicate that deep learning models, particularly LSTM, outperform conventional machine learning techniques in identifying both traditional and evasive spam messages while maintaining high classification accuracy and generalization capability. The proposed framework offers a scalable, reliable, and adaptable solution for intelligent SMS spam filtering and provides valuable insights into developing next-generation spam detection systems capable of addressing continuously evolving adversarial messaging techniques.