Modeling and Prediction of Hourly Ambient Ozone (O3) and Oxides of Nitrogen (NOx) Concentrations Using Artificial Neural Network and Decision Tree Algorithms for an Urban Intersection in India

The present study attempts to predict hourly ozone (O3) and oxides of nitrogen (NOx) concentrations near a traffic intersection in megacity Delhi, India, using artificial neural network (ANN) with the Levenberg–Maquardt (LM) algorithm and decision tree algorithms [e.g., reduced error pruning tree (REPTree) and M5 P tree model]. The hourly averages of input variables of meteorological, traffic volume, and transport emissions along with target values of monitored ambient air concentrations of O3 and oxides of nitrogen NOx were used for model development. The LM, REPTree, and M5 P algorithm models were developed by training, validation, and testing of input and target data. Statistical agreement between observed and predicted values is assessed by coefficient of correlation (CC), mean square error (MSE), root mean square error (RMSE), normalized mean square Error (NMSE), and Nash–Sutcliffe efficiency index (N-S Index). Results show that the performance of the M5 P model is superior to ANN and REPTree models studied for prediction of O3 and NOx at a highly urbanized traffic intersection.