Detecting smishing attacks on smartphones: a comparative study between supervised and unsupervised learning techniques

Abstract

Smishing is a phishing technique that involves sending short messages (SMS) to a smartphone user's inbox to defraud them and disclose their personal information using social engineering. This type of attack is often used for financial theft by attackers. The excessive multiplication of these attacks leads cyber security researchers to employ advanced detection techniques such as machine learning. Our work involves evaluating the detection performance of selected Machine Learning algorithms using supervised and unsupervised learning and comparing the results obtained regarding detection accuracy. For supervised learning, we chose to work with these classification algorithms, which gave the following results: Naive Bayes (96%), Decision Tree (94%), Random Forest (95%), and Multi-layer perceptron (92.2%). For unsupervised learning, we chose the Gaussian Mixture Model algorithm, which gave an overall accuracy of 59%. The results demonstrate that supervised learning models are better at detecting smishing attacks than unsupervised learning models, which require the parallel intervention of other analysis techniques, such as behavioral analysis, to improve the detection of such types of attacks.