Predicting air quality index using artificial neural network: a case study of the Coimbatore region

This study predicts the Air Quality Index (AQI) in the Coimbatore region using Artificial Neural Networks (ANN) to enhance air quality management and environmental science. The researchers measured six input variables over a two-year period, including Particulate Matter size less than 2.5 microns (PM_2.5), Particulate Matter size less than 10 microns (PM10), Sulphur dioxide (SO₂), Nitrogen oxide (NO), Carbon monoxide (CO), and Ozone (O₃). The AQI was then calculated to evaluate the impact of atmospheric contaminants. The dataset was divided into 70% for training and 30% for testing. Models were developed using two algorithms: the Levenberg-Marquardt (LM) algorithm and the Bayesian Regularization (BR) algorithm. Analysis of Variance (ANOVA) is adopted to examine the relationship between the variables. Additionally, sensitive assessments, including the coefficient of determination (R²), mean square error (MSE), and root mean square error (RMSE), were conducted to evaluate the influence of neurons, layers, and input variables. The LM algorithm outperformed the BR algorithm, achieving its best performance with a cross-validation as K = 9 for the sixth neuron. The optimal model yielded a high R² of 0.9604 and low error metrics, with an MSE of 0.00235 and an RMSE of 0.0485.