SMS-Shield: A Lightweight Approach for Smishing Detection using Machine Learning

SMS continues to be an indispensable component of contemporary digital communication. Because it is so widely available, cost-effective, reliable, and delivers information instantly. Its versatility enables its utilization for real-time chats, alerts, and verification purposes. Nevertheless, the ubiquitous nature of SMS renders it an easy prey for smishing, a deceitful form of cyberattack. Cybercriminals exploit the trust associated with SMS messages to deceive recipients. Smishing, an increasingly prevalent deceptive SMS fraud, poses a threat to both businesses and individuals. These messages include websites or phone numbers and purport to be from banks, governmental bodies, or well-known businesses. These misleading text messages lure their users into disclosing private information or clicking on links that could be harmful in many ways. Early detection of such malicious messages prevents users from falling victim to them. Various techniques were presented by researchers to accurately detect smishing attacks. Machine learning models exhibit great potential as a strategy to counter the increasingly dangerous SMS phishing attacks. Machine learning models can discern legitimate from fraudulent SMS messages by analyzing datasets. This paper introduces a novel method (SMS-SHIELD), which combines Machine Learning and Natural Language Processing techniques to enhance smishing detection. This approach yields outstanding results, with random forest and extreme gradient boosting algorithms achieving remarkable accuracy rates of 99.05% and 98.03%, coupled with impressive F1-scores of 98.95% and 97.86%, respectively.