Multi-layer perceptron model for air quality prediction

Abstract

This study trained two MLP models with dierent activation functions in assessing the capability of the model for the prediction of air quality. The daily air quality data and meteorological variables from year 2010-2014 were assembled in training and testing the models. The MLP model with the combination of tansig and purelin activation function revealed 69.0% of variance in data with 5.58 ηg/m3 (RMSE) and 80.0% of variance in data with 8.14 ηg/m3 (RMSE), during training and testing phase, respectively. This model is appropriate for operational used by respected authorities in managing air quality and as early warning during unhealthy level of air quality. © 2019, Universiti Putra Malaysia.