Air Pollution Forecasting using Hybrid MLR-RNN Method with Time-Stratified Method

Abstract

studying and forecasting Particular matter (PM10) is necessary to control and reduce the damage of environment and human health. There are many pollutants as sources of air pollution may effect on PM10 variable. This type of dataset can be classified as anonlinear. Studied datasets have been taken from climate station in Malaysia. Multiple Linear Regression (MLR) is used as alinear statistical method for PM10 forecasting through its influencing by corresponding climate variables, therefore it may reflect inaccurate results when used with nonlinear datasets. Time stratified (TS) method in different styles is implemental for satisfying more homogeneity of datasets. It includes ordering similar seasons in different years together to formulate anew variable smoother than their original. To improve the results of forecasting, Recurrent Neural Network (RNN) has been suggested to be used after combining with MLR in hybrid MLR-RNN method in this study. In general, the results of forecasting were the best with using time stratified approach. In addition, the results of hybrid method were outperformed comparing to MLR model. As conclusion in this study, RNN and TS can be used as active approaches to obtain better forecasting results with nonlinear datasets in which PM10 is to dependent variable.