SMS SPAM DETECTION USING HYBRID DEEP LEARNING

Abstract

The unprecedented increase of spam and promotional SMS messages has created the new challenge of quickly and accurately detecting these messages in the field of information security and communication. Users lose time and money to these spam messages, and these spam messages can also result in even greater financial and security risks. Therefore, building systems capable of detecting messages and analyzing their various features is extremely valuable.To combat the spam SMS phenomenon, this paper presents the first of its kind Hybrid Deep Learning Architecture, comprised of One-Dimensional Convolutional Neural Networks (CNNs), Bidirectional Long Short-Term Memory (BiLSTM), and Transformer encoder blocks for SMS spam detection. The architecture is designed to accurately capture local features, temporal dependencies, and long-range relationships to classify messages into ham and spam. Preprocessing such as text cleaning and tokenization, padding, and balancing classes were done to the dataset in preparation for training the model. The model demonstrates substantial successful performance to all previous studies within the domain. Experimental results demonstrate that the proposed model achieves outstanding performance with 99.9% accuracy, 100% precision, 99.8% recall, 99.9% F1-score, and AUC = 1, outperforming previous methods. These findings highlight the potential of the proposed hybrid architecture as an effective and practical solution for SMS filtering in real-world applications.