Toward Explainable Digital Twins for Indoor Environmental Risk Prediction Using Open IoT Sensor Data

Although digital twins have become a viable technology for intelligent building management, the majority of current applications prioritize real-time monitoring and visualization above proactive environmental risk prediction. In order to estimate future interior environmental threats utilizing multimodal Internet of Things (IoT) sensor data, this study suggests an explainable digital twin paradigm. We used an open-access indoor environmental dataset that included about 8,000 observations from five sensor nodes. Temperature, air quality, sound, lighting, proximity, and temporal measures were all included in the dataset. Lagged variables, rolling-window statistics, interaction features, and contextual indicators were produced using a multimodal feature-engineering technique. For the purpose of predicting future overheating risk, five machine learning algorithms (Logistic Regression, Random Forest, XGBoost, LightGBM, and Gradient Boosting) were assessed. Uniform Manifold Approximation and Projection (UMAP) and KMeans clustering were used to find latent environmental risk states, and Explainable Artificial Intelligence (XAI) was integrated with SHapley Additive exPlanations (SHAP) to identify the environmental drivers of risk. All models showed outstanding predictive performance, with Random Forest attaining the best accuracy (0.980), F1-score (0.980), and AUC-ROC (0.998). While air quality, auditory, and temporal variables offered further predictive information, temperature was found to be the most significant predictor of future risk. Additionally, the clustering methodology aided the understanding and decision-making of environmental states by identifying different low-, moderate-, and high-risk operational situations. The suggested paradigm transforms digital twins from passive monitoring systems into transparent, intelligent, and predictive platforms that can help with risk reduction and proactive indoor environmental management.