Title : Smart cities, cleaner skies: An explainable AI framework for urban air quality prediction and environmental decision-making
Abstract:
Urban air pollution remains a persistent challenge for smart-city governance, as meteorological variability, microclimatic conditions and anthropogenic activities jointly drive fluctuations in fine particulate matter (PM?.?) concentrations. Although machine-learning models offer strong predictive capability, their black-box mechanisms limit transparency and reduce stakeholder confidence, particularly when forecasts inform public-health advisories and environmental planning. This study proposes an explainable artificial intelligence (XAI) framework for multi-horizon urban PM?.? forecasting and environmental decision support, using the Klang Valley, Malaysia, as an empirical case study. The framework integrates historical PM?.? observations, temporal lag features and meteorological variables, including temperature, relative humidity, surface pressure, wind speed, wind direction and rainfall. Random Forest and Extreme Gradient Boosting (XGBoost) models are benchmarked against a multiple linear regression baseline using chronological time-series validation, which prevents information leakage from future observations. Forecasting performance is evaluated using mean absolute error (MAE), root mean squared error (RMSE) and the coefficient of determination (R²). To open the black box, SHapley Additive exPlanations (SHAP) are applied to interpret global feature importance and individual predictions. Explanation faithfulness and stability are also assessed, while atmospheric domain knowledge informs the evaluation of physical plausibility, for example whether the attributed influence of wind speed and rainfall is consistent with dispersion and wet-scavenging processes. To evaluate practical value, the study is grounded in the DeLone and McLean Information Systems Success Model. Survey data from environmental professionals and potential decision-support users are analysed using Partial Least Squares Structural Equation Modelling (PLS-SEM) in SmartPLS. The analysis examines how perceived system quality and information quality influence user satisfaction and perceived decision-support benefits, linking predictive performance and explanation reliability with users’ assessments of system usefulness. The study aims to contribute an empirically evaluated framework that combines accurate forecasting, interpretable predictions and user-centred system evaluation. By connecting technical performance with stakeholder perceptions, it supports transparent urban air-quality monitoring and evidence-informed environmental planning in the Klang Valley, and offers a potentially transferable approach for other rapidly urbanising regions.
Keywords: smart cities; urban air quality; PM?.?; explainable artificial intelligence; SHAP; DeLone and McLean Information Systems Success Model; SmartPLS
