Robust urban air quality index prediction for pollution management via entropy-guided multiscale denoising and optimization-driven neuro-fuzzy modeling

Urban air quality is a critical determinant of environmental sustainability and public health, directly influencing pollution management and climate resilience. Accurate prediction of the urban air quality index (AQI) plays a pivotal role in guiding policy decisions, mitigating health risks, and supporting effective environmental governance. However, the strong non-stationarity and complex noise characteristics inherent in AQI time series substantially hinder reliable modeling. To address these challenges, this study proposes a robust forecasting framework that integrates entropy-guided multiscale signal denoising with an optimization-driven adaptive neuro-fuzzy inference system (ANFIS). Specifically, an enhanced quadratic interpolation-based hiking optimization algorithm (QIHOA) is developed by embedding a quadratic interpolation scheme with a dynamic alert mechanism, thereby improving global search capability and maintaining population diversity. In parallel, a multiscale incremental entropy (MIE)-based hierarchical denoising method is constructed, which combines improved complete ensemble empirical mode decomposition with adaptive noise (ICEEMDAN) and MIE-based clustering to isolate high-frequency noise components, further refined through variational mode decomposition (VMD) for noise suppression. Subsequently, a parameter-adaptive QIHOA-optimized ANFIS model is designed to alleviate sensitivity to parameter settings and enhance prediction accuracy. Extensive experiments on AQI datasets from three major Chinese cities demonstrate that the proposed framework achieves superior forecasting accuracy and robust generalization. • Proposed Interpolation and alert-based QIHOA enhances convergence and exploration. • MIE-clustered denoising method with ICEEMDAN and VMD for robust noise reduction. • Parameter-adaptive QIHOA-ANFIS improves reliability and accuracy in AQI prediction. • Proposed framework shows strong performance and generalization across major cities.