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

COMPUTER SCIENCE AND ENGINEERING

In the digital age, SMS remains a popular medium for communication, yet it is increasingly exploited by spammers who employ sophisticated evasion techniques to bypass traditional spam filters. These techniques include deliberate obfuscation of words, the use of special characters, and mimicry of legitimate content, all of which challenge the effectiveness of conventional filtering systems. This study investigates the impact of such evasive tactics on spam detection and evaluates the performance of various machine learning models in identifying and classifying spam messages. A comprehensive dataset comprising both standard and obfuscated spam messages is used to train and test multiple machine learning classifiers, including Naive Bayes, Support Vector Machines (SVM), Decision Trees, Random Forests, and deep learning models such as Long Short-Term Memory (LSTM) networks. Preprocessing techniques such as tokenization, stop-word removal, stemming, and vectorization are applied to enhance model accuracy. The results of the comparative analysis reveal that while traditional models perform well on regular spam, their accuracy declines when faced with obfuscated or evasive messages. In contrast, advanced models like LSTM demonstrate greater resilience due to their ability to capture contextual and sequential dependencies in text. The study emphasizes the need for adaptive and intelligent spam filtering solutions to counter the evolving strategies of SMS spammers.

This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.