ARTIFICIAL NEURAL NETWORK MODEL FOR AIR POLLUTION FORECASTING IN KADUNA, NIGERIA

Abstract

The goal of air quality forecasting is to predict when air pollution concentrations will reach levels that are unsafe for human health. There are significant regional differences in air quality, hence a generalized forecast model is not effective. It is necessary to develop localized forecast models. It is challenging to identify a more accurate forecast model for any given environment. This work produces an accurate forecast model for the area under study. In this study, air pollution data was acquired from the three different sampling stations in Kaduna, Nigeria. The data was used to train Artificial Neural Network (ANN) models for each of the sampling stations. These models were implemented using feed forward backpropagation (FFBP) algorithm. The simulations of FFBP were performed with a varying number of neurons in the hidden layer. The resulting models were used to forecast the next ten days for each of the sampling station and for each pollutant. Determination of the accuracies of the developed models in forecasting the next ten days was achieved using error performance metrics of Mean Absolute Error (MAE), Root Mean Squared Error (RMSE) and Mean Absolute Percentage Error (MAPE). The results of the performance metrics from most of the models in the same category are correlated and indicate similar trends. Comparison and analysis of the models revealed the model with the most accurate prediction for each sampling station and pollutant.