Air Pollution Forecasting using Fuzzy Time Series Models for Kaduna Metropolis, Nigeria

Abstract

Fuzzy Time Series (FTS) is able to eliminate the problem of overfitting that is fundamental to Artificial Neural Network (ANN), hence this study used air pollution data acquired from three different sampling stations in Kaduna metropolis, Nigeria, to implement FTS using the Adaptive Neuro Fuzzy Inference System (ANFIS). The fuzzy inference system (FIS) was generated by the ANFIS model using grid partitioning and subtractive clustering optimization types with backpropagation and hybrid training algorithms. The models were implemented using MATLAB 2018b software, and a total of thirteen models were developed. The resulting models were used to forecast the daily mean for the next ten days for each sampling station and for each pollutant. Carbon monoxide (CO), Nitrogen dioxide (NO2), Sulfur dioxide (SO2), Particulate matter, (PM2.5 and PM10) air pollutants were considered. Determination of the accuracies of the developed models in forecasting the next ten days was achieved using the error performance metrics of Mean Absolute Error (MAE), and Root Mean Squared Error (RMSE). The results of the performance metrics from the models in the same category are correlated and indicated similar trends. Comparison and analysis of the models revealed the one with the most accurate prediction for each sampling station and pollutant.