A Multi-Task Machine Learning Approach for Joint SMS and URL Safety Classification

Abstract

 The sudden evaluation of SMS communication has rendered mobile consumers more susceptible to spam and phishing threats, particularly across embedded URLs found in both unsolicited and authentic texts. Current methods failing to particularly assess the safety of the embedded URLs, focusing on SMS text as either spam or ham, so allowing hidden threats to remain invisible. This article introduces a dual-task architecture that URL safety classification on a managed SMS dataset, and also concurrently executes SMS spam detection and wherein each message includes at least one URL. We apply preprocessing technique in SMS text utilizing natural language processing approaches and extract SMS features, lexical, structural, and reputational information from cleaned SMS Dataset to train separate yet collaborative models. When SMS is marked as authentic, The SMS level classifier employs text-based features to differentiate between spam and ham, whereas the URL level classifier autonomously assesses the safety of each link. This article is displayed on machine learning approaches like support vector machine, Logistic regression and Naïve Bayes, which algorithm provides the best recall, accuracy, precision, and F1 score for a real-world SMS spam dataset enhanced with URL annotations demonstrate that the both tasks individually and combined for mobile messaging services. cyber security mechanism that supports more comprehensive real-time protection against phishing and spam attacks in SMS mobile services.