A New Machine Learning Framework for Smishing Detection Using Advanced Text Normalization

The proliferation of smartphones has made SMS-based phishing, or "smishing," a significant and pervasive cybersecurity threat. Traditional detection methods often fail to keep pace with the evolving and informal nature of SMS language, which is characterized by slang, abbreviations, and intentional misspellings. This linguistic noise fragments the feature space for machine learning classifiers, leading to reduced accuracy and higher false positive rates. This paper introduces a novel, two-stage machine learning framework designed specifically to address this challenge. The core innovation is the integration of an advanced text normalization engine as a critical preprocessing step before classification with a Naive Bayes model. The normalization module systematically converts non-standard SMS language into standardized English using a curated Lingo Dictionary, thereby consolidating the feature space and enhancing the semantic clarity of the message content. We conduct a rigorous empirical evaluation on a curated dataset derived from the public SMS Spam Collection. The results demonstrate that our framework achieves a detection accuracy of 96.2%, a True Positive Rate of 97.14%, and a False Positive Rate of just 3.87%. This represents a substantial improvement over a baseline model without normalization and outperforms several existing state-of-the-art approaches. The findings confirm that explicitly addressing linguistic variability is essential for building robust and effective smishing detection systems.