Forecasting PM10 Concentrations Using Artificial Neural Network in Imphal City

Abstract

In this study, a forecasting system is developed for predicting PM10 levels in Imphal City over the next three days (+1, +2, and +3 days) using artificial neural networks (ANN). The experimental findings indicate that the ANN model can achieve reasonably accurate predictions of air pollutant levels. Moreover, optimizations in model performance are explored through variations in input parameters and experimental setups. Initially, predictions for each of the +1, +2, and +3 days are made independently using the same training dataset. Subsequently, cumulative predictions for +2 and +3 days are generated using previously predicted values from preceding days, yielding improved prediction accuracy. Additionally, the study identifies the optimal size of the training dataset, determining that using data spanning 3 to 15 past days yields the minimum error rates in predicting pollutant concentrations. Finally, the investigation includes the consideration of days-of-week as an input parameter, which enhances forecast accuracy noticeably.