Comparative Analysis of Machine Learning Models for Smishing Message Detection

Aqsa Shaikh, Mariya Shaikh, Srivaramangai R

Abstract: The advent of mobile messaging at a very fast pace has brought about the emergence of smishing (SMS phishing) as a major cybersecurity challenge. The perpetrators of the crime take advantage of SMS messages to deceive the victims into revealing their credentials or clicking on harmful links. The paper elaborates on the examination of five machine learning models - Logistic Regression, Random Forest, Support Vector Machine (SVM), Convolutional Neural Network (CNN), and Long Short-Term Memory (LSTM) - in terms of their ability to correctly classify smishing messages. The researchers utilize the widely available UCI SMS Spam Collection dataset and perform text preprocessing and feature extraction through TF-IDF and word embedding. The study also assesses the performance of each model based on the same standard classification metrics. The experimental results indicate that deep learning models have better performance than traditional techniques, with LSTM providing the best detection accuracy of 98.2%. Ensemble methods like Random Forest have also shown to be very accurate and interpretable, thus presenting a balanced performance which is appropriate for real-time applications.

Keywords: Smishing Detection, Machine Learning Models, Deep Learning, SMS Spam Classification, TF-IDF Feature Extraction, LSTM Network

How to Cite?: Aqsa Shaikh, Mariya Shaikh, Srivaramangai R, "Comparative Analysis of Machine Learning Models for Smishing Message Detection", Volume 14 Issue 12, December 2025, International Journal of Science and Research (IJSR), Pages: 1984-1990, https://www.ijsr.net/getabstract.php?paperid=SR251223103916, DOI: https://dx.doi.org/10.21275/SR251223103916