ARCHIVES

The rapid proliferation of mobile devices has escalated the sophistication of SMS-based spam and phishing attacks, posing severe cybersecurity threats. Traditional machine learning models often fail to capture the sequential dependencies and contextual nuances of short, noisy text messages. This paper presents a novel hybrid deep learning architecture, combining Convolutional Neural Networks (CNN) and Bidirectional Long Short-Term Memory (BiLSTM) networks, for highly accurate SMS spam detection. The CNN layer extracts prominent local features, such as suspicious URLs and urgent keywords, while the BiLSTM layer preserves the bidirectional sequential context of the message. To address the “black-box” nature of deep learning, we integrate Explainable AI (XAI) techniques to provide human-readable rationales for predictions, highlighting specific risk-inducing keywords. Furthermore, we deploy this model through a real-time Flask API integrated with a Java-based Android application, enabling on-device SMS scanning with minimal latency. Experimental evaluation on a combined dataset (UCI SMS Spam Collection and contemporary phishing datasets) demonstrates that our proposed CNN-BiLSTM model achieves an exceptional accuracy of 97%, significantly outperforming baseline algorithms like Naive Bayes, SVM, and standalone recurrent networks. The integration of high accuracy, real-time mobile deployment, and explainable predictions establishes a robust defense mechanism against modern mobile-based social engineering attacks