A Lightweight and Robust SMS Spam Filtering Model for Mobile Networks

Abstract

Mobile messaging has skyrocketed with the proliferation of mobile users, which has brought about an upsurge in SMS (Short Message Service) spam. Unwanted and sometimes dangerous spam text messages are a major obstacle to mobile communication. Using the 5,574 tagged messages (ham or spam) from the SMS Spam Collection dataset, this study aims to detect spam via machine learning. Preprocessing the raw text data involves stemming, removing special characters, lowercasing, and tokenizing. Then, features are extracted using TF-IDF and dimensionality is reduced using Principal Component Analysis (PCA). A few performance metrics are utilized to assess KNN and NB, two classification models, including accuracy, precision, recall, F1-score, ROC curve, and confusion matrix. When it came to spam detection, KNN had the best accuracy (95.3% and 98.5%, respectively), whereas NB was the best in precision (97.6%, minimizing false positives). Examining KNN and NB alongside other classifiers, like Decision Tree (DT) and Random Forest (RF), reveals that they outperform them. The study concludes that both models are highly effective for SMS spam filtering, with KNN preferred in high-recall applications and NB in precision-critical scenarios, making them suitable for mobile and resource-constrained environments.