Given the escalating focus on the environment by governmental bodies and the persistent decline in air quality, this study advocates an upgraded version of the adaptive neuro-fuzzy inference system (ANFIS) for predicting the air pollutant index in Mumbai, India. The proposed methodology, which is the ensemble of covariance matrix learning differential evolution and Adam (CoDEAdam), amalgamates the modified Differential Evolution (DE) algorithm with hybrid Adam, a contemporary optimization technique. CoDEAdam employs two distinct sub-algorithms to evolve two sub-populations concurrently, enabling effective exploration of global and local search spaces and ensuring rapid convergence. To train the model, nine years of historical air pollutant index data from Mumbai city are utilized, forecasting pollutants such as particulate matter (PM10), Sulfur dioxide (SO2), and carbon monoxide (CO) due to their significant impact on Mumbai’s air quality. A comparative analysis with DE-ANFIS, PSO-ANFIS and decision tree-based models reveals substantial enhancements in performance achieved by the proposed CoDEAdam-ANFIS approach.