Interpretable AI Models for IoT-Enabled Environmental Monitoring and Disaster Risk Forecasting

The increasing frequency and severity of environmental disasters necessitate advanced methodologies to predict and mitigate their risks effectively. This paper introduces an interpretable AI framework tailored for forecasting environmental disaster risks using data from IoT-enabled sensors distributed across diverse geographies. The study leverages a comprehensive dataset containing key climate, soil, and air parameters sampled hourly over six months. Our methodology encompasses the use of interpretable machine learning models such as SHAP and LIME, which elucidate predictions by providing feature attribution, thus enhancing transparency in decision-making processes crucial during high-risk scenarios. Results indicate that incorporating explainability mechanisms significantly improves model trustworthiness without compromising prediction accuracy. Notably, the proposed models achieve high metrics, including accuracy, macro f1-score, and consistency of SHAP scores across trials. These findings underscore the potential of explainable AI to bolster environmental monitoring systems, ensuring informed decisions that can mitigate disaster impacts effectively while fostering trust among stakeholders.