Prediction of Air Pollution Index in Kuala Lumpur using fuzzy time series and statistical models

• Published in: Air quality, atmosphere and health Vol. 13; no. 1; pp. 77 - 88
• Main Authors: Koo, Jian Wei, Wong, Shin Wee, Selvachandran, Ganeshsree, Long, Hoang Viet, Son, Le Hoang
• Format: Journal Article
• Language: English
• Published: Dordrecht Springer Netherlands 2020; Springer Nature B.V
• Subjects: Air pollution; Air pollution forecasting; Air quality; Air quality forecasting; Artificial neural networks; Atmospheric Protection/Air Quality Control/Air Pollution; Autoregressive models; Comparative studies; Computer applications; Computing time; Earth and Environmental Science; Environment; Environmental Health; Forecasting; Forecasting techniques; Health Promotion and Disease Prevention; Immunology; Mathematical models; Model accuracy; Neural networks; Outdoor air quality; Performance evaluation; Pollutants; Pollution index; Respiratory diseases; Root-mean-square errors; Seasonal variations; Statistical analysis; Statistical models; Statistical prediction; Time series; Kuala Lumpur Malaysia; Malaysia; ARMA errors; Fuzzy logic; Fuzzy time series; Artificial neural network (ANN); Autoregressive integrated moving average (ARIMA); Trigonometric regressors; Box-Cox transformation; Trend and seasonality (TBATS) model; Air Pollution Index (API)
• ISSN: 1873-9318; 1873-9326
• DOI: 10.1007/s11869-019-00772-y

Air pollutants can cause multifaceted harm to the human body. Respiratory diseases and immunology dysfunction are some of its main manifestations. Forecasting the air quality of a country is important to allow the government to take preventive measure. In this research, the artificial neural network (ANN), autoregressive integrated moving average (ARIMA), trigonometric regressors, Box-Cox transformation, ARMA errors, trend and seasonality (TBATS) and several fuzzy time series (FTS) models are utilized in the forecasting of air pollution index (API) of Kuala Lumpur, Malaysia, for the year 2017. Six years of daily API data for Kuala Lumpur from the year 2012 to the year 2017 for the Cheras observation station in Kuala Lumpur has been selected as the dataset of this research. The mean absolute percentage error (MAPE), root mean square error (RMSE) and computational time have been used as the performance evaluation metrics for the models and these values were calculated for each of the chosen forecasting models. A brief but comprehensive comparative study of the results obtained from each of the chosen model is presented in order to identify the most effective model to forecast API values. It was found that the fuzzy time series models outperformed the other models in terms of accuracy of forecasted values and computation time. Specifically, the Singh fuzzy time series model was found to be the most accurate and efficient forecasting model with RMSE of 1.4704 and MAPE of 4.364%.