Investigating Evasive Techniques in SMS Spam Filtering: A Comparative Analysis of Machine Learning Models

Abstract: The proliferation of SMS spam poses a growing threat in the domain of cybersecurity, often resulting in financial scams, privacy breaches, and negative user experiences. Although machine learning algorithms have become central to spam detection, adversaries continue to refine their evasion tactics—employing methods like character obfuscation, altered vocabulary, and adversarial perturbations. These evolving techniques challenge the effectiveness of traditional rule-based and standard machine learning systems, highlighting the demand for more resilient and adaptive spam-filtering frameworks. This review delivers a thorough exploration of modern evasion strategies in SMS spam and critically evaluates the performance of machine learning and deep learning models in detecting such threats. We investigate conventional classifiers—such as Naïve Bayes, Support Vector Machines (SVM), and Decision Trees—alongside cutting-edge deep learning models like Recurrent Neural Networks (RNN), Long Short-Term Memory (LSTM), and Transformer architectures including BERT and GPT. Additionally, ensemble learning approaches that combine the strengths of multiple models are discussed to assess their contribution to improved detection accuracy. To support ongoing research, we spotlight key challenges that remain unaddressed in SMS spam filtering, such as the demand for real-time and privacy-preserving solutions, and language-independent filtering mechanisms. We also emphasize the necessity for ethically grounded AI systems that can minimize inherent biases in spam classification. Keywords: SMS spam, evasion strategies, machine learning, deep learning, adversarial examples, NLP, robust classifiers, spam filtering models