A hybrid content and character feature method for SMS spam detection

Abstract

Short message service (SMS) spam remains a significant challenge due to its impact on user security and communication efficiency. This study proposes a hybrid spam detection model, convolutional neural network with content and character-based features (CNN-CCB), which integrates word-level features and character-level features, with handcrafted content-based and character-based features in a unified deep learning framework. Unlike conventional CNN and hybrid models that rely primarily on learned representations, the proposed approach incorporates structural features to enhance detection of short and noisy text patterns. Firstly, the text data were tokenized and processed through dual convolutional branches, while handcrafted features are fused to improve classification. Secondly, class weighting is applied to address data imbalance while maintaining predictive reliability. Finally, regularization techniques are employed to prevent overfitting. The experimental results show that CNN-CCB achieves high performance, with an accuracy of 0.997, precision of 0.988, recall of 0.998, and F1-score of 0.993, outperforming baseline models such as long short-term memory (LSTM) dan gated recurrent unit (GRU). The model demonstrates consistent performance across three datasets, indicating the model’s robustness and generalizability. These findings suggest that the proposed hybrid framework is effective for SMS spam detection and has potential applications in mobile security and real-time communication systems.