Machine Learning Approach for the Classification of Short Message Service Spam

Abstract

The unwanted advertisements sent with Short Message Services (SMS) significantly interferes the normal communication among customers, which should have been guaranteed by the mobile service providers. Such trends have attracted many phishing and spamming attacks using SMS messages. The issue with SMS spam classification lies in its direct impact on user security, privacy, communication experience, technological advancements, and the broader social and economic landscape. It plays a crucial role in fostering a safer and more efficient digital environment for users. They can unknowingly access suspect websites and be at risk of phishing attacks or malware downloads. The study embarked on an extensive exploration of SMS spam classification techniques, aiming to distinguish between legitimate messages (ham) and unsolicited or unwanted messages (spam). Kaggle database provided us the dataset “SMS Spam Collection” which is made up of over 10,000 records of Ham and Spam messages. The “SMS Spam Collection” dataset is a public set of SMS labelled messages that have been collected for mobile phone spam research. The dataset was splited into 70% training set and 30% testing set. The performance was evaluated based on Accuracy, Precision, Recall and F1-Score. Based on the analysis of the experimental results of the NB-SVM hybridized model, the Accuracy of 0.98 (98%), 0.99 and 0.91 Precisions for HAM and SPAM respectively, 0.99 and 0.93 Recall for HAM and SPAM respectively, and 0.99 and 0.92 F1-Score for HAM and SPAM respectively.