Air pollution is a global problem with an immense negative impact on the health of humans and the climate. The Air Quality Index (AQI) is a very effective measure for quantifying the severity of air pollution, and highly accurate forecasting ability is required to make it effectively applicable to early intervention strategies. In this study, we introduce a hybrid approach to forecasting Air Quality Index (AQI) using Variational Mode Decomposition (VMD) combined with eighteen different predictive models—including state-of-art machine learning, deep learning, and stacked ensemble models. To evaluate the performance of each model, we considered four widely used metrics: SMAPE, RMSE, MASE, and MAE. Among these, the VMD-based Stacked-Huber regression model consistently outperformed all other models and secures the best overall accuracy. The Proposed method demonstrates significantly improved performance compared to other individual models, with an improvement of 18.7% in SMPAE, 16.2% in RMSE, and 21,4% in MAE. In order to verify the robustness of these findings, two statistical tests namely, Wilcoxon Signed Ranked test and Friedman Nemenyi Hypothesis test were conducted. The findings indicate that the VMD based Stacked-Huber regression secures better and reliable results for daily AQI forecasting.