Air pollution has now become one of the major ecological problems in the Indian cities that has several negative effects on human health and environment. This study developed and tested an Artificial Neural Network (ANN) to predict multi-pollutants in five cities of Madhya Pradesh, i.e. Bhopal, Dewas, Indore, Jabalpur and Ujjain, to examine the city-specific pollutant concentration and meteorology relationships. The preprocessing of five years of hourly CAAQMS data (2019 to 2023) of PM10, PM2.5, NO2, and SO2 with meteorological parameters (temperature, humidity, wind speed, wind direction) of a single monitoring station per city was done using weighted bidirectional forward fill imputation and min-max normalization. The neural network architecture consisted of four hidden layers (512-256-128-64 neurons) with ReLU activation function and trained on Adam optimizer with 80:25 train-test split. The model showed strong performance for PM₁₀, PM₂.₅, and NO₂ (MSE: 0.13-0.39, R²: 0.65-0.87). Jabalpur and Indore showed better accuracy (R² > 0.80) due to stronger pollutant-meteorology correlations and stable industrial emission patterns. SO₂ predictions were weaker (R²~0.50-0.67) due to dispersed combustion sources in study and complex atmospheric chemistry. This study provides the first comprehensive multi city ANN validation for central Indian non-attainment cities. The next step involves the following: the expansion of this methodology to other cities, the improvement of SO2 prediction with the help of hybrid models, and the application of the constructed model to the real-time working forecasting systems.
