Smishing Detection and Domain Identification Using Deep Residual Fuzzy Encoders for Multi-Modal Classification

Abstract

SMS phishing (smishing) attacks have become a growing concern in the era of ubiquitous mobile communication. Detecting smishing messages requires understanding not only the malicious content but also the context and origin of the messages. This research paper introduces a novel method for smishing detection that combines a domain identification phase with spam classification using deep residual fuzzy encoders (DRFE) for multi-modal classification. The proposed methodology has two essential components. First, a domain identification phase is implemented to distinguish between legitimate and suspect domains associated with the SMS sender. This phase employs machine learning techniques to analyze the domain's attributes, such as its reputation, content, and historical patterns. By identifying the domain, the detection system gains valuable context that aids the classification phase that follows. Second, for spam classification, a deep residual fuzzy encoder (DRFE) model is employed. The DRFE model combines deep learning architectures and fuzzy logic in order to capture the complex patterns and inherent uncertainty in smishing messages. In the presence of noise or ambiguity in the data, the model can effectively extract informative features and make accurate predictions by combining residual connections and fuzzy inference. The proposed approach employs multimodal classification techniques to address the multimodal nature of smishing messages. The textual content of the messages is combined, if possible, with additional features extracted from accompanying images. This integration permits the model to utilize both textual and visual cues, thereby improving the overall accuracy and robustness. Experimental results demonstrate that the proposed method outperforms baseline approaches in domain identification and spam classification. The incorporation of deep residual fuzzy encoders and multi-modal classification significantly improves the detection accuracy, making it a promising solution for defending against smishing attacks in real-world situations.