Explainable AI for SMS spam filtering: A novel hybrid architecture combining fuzzy logic and bidirectional LSTM networks

SMS spam filtering has become a critical application in modern telecommunications, where unsolicited and fraudulent text messages compromise user security, erode trust, and burden network resources. To address these challenges, this study proposes a novel hybrid architecture that integrates fuzzy logic with a bidirectional long short-term memory (BiLSTM) network to achieve high classification performance and transparent decision-making. The fuzzy logic module captures linguistic uncertainty using Gaussian membership functions, while the BiLSTM effectively models sequential dependencies in text. To further enhance interpretability, token-level attention highlights influential words, and rule-based explanations provide human-understandable justifications for each classification outcome. The proposed framework was evaluated on two benchmark datasets: the UCI SMS Spam Collection and the ExAIS_SMS dataset. Experimental results demonstrate strong performance, with accuracies of 99.4 % and 95.8 %, respectively, outperforming several baseline approaches. Notably, the model also achieves high discriminative capability, with AUC values of 0.994 and 0.980 across the two datasets. Beyond accuracy, the framework is computationally efficient, requiring <2.5 GB of memory and averaging under 7 ms per message during inference on a consumer-grade GPU. This ensures feasibility for real-time, large-scale deployment. By balancing accuracy, interpretability, and efficiency, the proposed fuzzy-deep hybrid approach offers a practical and trustworthy solution for SMS spam detection in telecom environments.