Improving Air Pollution Prediction Modelling Using Wrapper Feature Selection

Feature selection is considered as one of the essential steps in data pre-processing. However, all previous studies on predicting PM10 concentration Malaysia have been limited to statistical method feature selection, and none these used machine-learning approaches. Therefore, objective this research investigate influence variables prediction model by using wrapper compare performance different predict for next day. This uses 10 years daily pollutant concentrations from two stations (Klang Shah Alam) obtained Department Environment (DOE) 2009 until 2018. Six methods (forward backward elimination, stepwise, brute-force, weight-guided genetic algorithm evolution predictive analytics multiple linear regression (MLR) artificial neural network (ANN)) were implemented study. study found that brute-force dominant most best models selecting important features MLR. Moreover, compared MLR, ANN provides more advantages regarding accuracy permits PM10. The overall results revealed RMSE value day Klang 20.728, while AE 15.69. Furthermore, Alam 10.004, 7.982. Finally, predicted can be concentrations. proposed a tool an early warning system giving air quality information local authorities order formulate air-quality-improvement strategies.