An Adaptive End-to-End IoT Security Framework Using Explainable AI and LLMs

Explainability makes security decisions understandable to humans: The framework uses SHAP and LIME techniques to explain why the system flagged specific activities as threats, moving beyond "black box" AI systems that security teams cannot interpret or trust.

Large Language Models translate technical alerts into actionable guidance: By incorporating LLMs like Gemini and OpenAI, the system converts complex security detections into plain-language explanations and specific mitigation strategies that administrators can actually implement, rather than just receiving alerts.

Different AI tools have complementary strengths: The research found that Gemini excels at providing precise, focused security responses while OpenAI offers more comprehensive, in-depth protective measures—suggesting that responsible AI deployment may benefit from combining multiple tools rather than relying on a single solution.

The framework bridges the gap between research and real-world deployment: Unlike many academic papers, this work demonstrates an end-to-end system that can move from development to actual deployment, addressing a practical gap where security innovations often fail to reach production environments.

Interpretability enables continuous improvement: By using XAI techniques to provide insights into how the detection model works, the framework creates opportunities to identify and fix model weaknesses, supporting more trustworthy and effective security systems over time.