INVESTIGATING EVASIVE TECHNIQUES IN SMS SPAM FILTERING: A COMPARATIVE ANALYSIS OF MACHINE LEARNING MODELS

Abstract

The permanence of SMS spam still presents serious problems, so it is necessary to create efficient detection systems that can manage the increasingly complex evasion tactics used by spammers. By offering a thorough SMS spam filtering system [14] that makes use of machine learning models—with a particular emphasis on Long Short-Term Memory (LSTM) networks—this study tackles these issues. We provide a new SMS dataset message that is the largest publicly available SMS spam dataset to date, consisting of 39% spam and 61% real (ham) messages. The evolution of spam was analysed longitudinally, and then syntactic and semantic aspects were extracted for assessment. Next, we compared several machine learning techniques, from simple models to sophisticated deep neural networks. Our research shows that conventional anti-spam services and shallow models are susceptible to evasion tactics, which leads to subpar performance. The LSTM model, on the other hand, performed better, classifying SMS messages with 98% accuracy. Even with this high accuracy, some evasion techniques continue to pose problems for identification, indicating areas that require more study. This study offers important insights into the efficacy of deep learning models in preventing SMS spam and promotes on-going improvement in reliable SMS spam filtering systems.