XAI-based Feature Selection for SMS Spam Classification in Dravidian Languages

With the exponential growth of mobile communication, the prevalence of SMS spam has become a significant concern, necessitating robust filtering mechanisms. This research focuses on optimizing SMS spam filtering in mobile applications by selecting the refined set of features for the two multilingual SMS Spam datasets. This paper uses eXplainable AI (XAI) approach for selecting the significant feature set based on the SHapley Additive exPlanations (SHAP) values. The average of SHAP values is used for setting the threshold value for the dataset. The feature whose feature importance score is greater than the threshold value is included in the significant feature set. Our experimentation demonstrates that employing fewer but highly significant features enhances the efficiency and accuracy of real-time SMS spam detection 90-92%. By leveraging this approach, mobile applications can deploy more resource-efficient and responsive spam filtering processes, thereby improving the overall user experience and security. This research contributes to the ongoing efforts in mitigating the impact of SMS spam, providing a valuable framework for enhancing spam filtering mechanisms in the dynamic world.