Bright Duffour, Alberta Ayitey
Misinformation detection has become increasingly critical in modern digital ecosystems, yet many existing approaches rely on complex black box models that limit interpretability. This paper presents Automated Explainable Guard for Information Security-Misinformation Identification System (AEGIS-MIS), a hybrid explainable misinformation detection system that integrates rule-based linguistic analysis with lightweight machine learning techniques. The system employs Term Frequency-Inverse Document Frequency feature extraction and classical classifiers, including Logistic Regression and Support Vector Machines, combined with an explainability module that provides transparent reasoning for each prediction. The proposed framework is evaluated across both controlled synthetic data and a real-world benchmark derived from the LIAR dataset. While strong performance is observed in structured prototype settings, results on benchmark data demonstrate moderate but realistic performance, highlighting the inherent challenges of misinformation classification. Comparative experiments show that model selection and feature engineering improve performance consistency, with Support Vector Machines achieving stronger class-level detection and feature union approaches enhancing robustness. The system is deployed as a lightweight web application and application programming interface, demonstrating practical usability alongside interpretability. The findings emphasize the importance of combining rule-based reasoning with machine learning inference, positioning AEGIS-MIS as an extensible and transparent framework for real-world misinformation detection.