Hybrid Vector Autoregression Feedforward Neural Network with Genetic Algorithm Model for Forecasting Space-Time Pollution Data
Rezzy Eko Caraka, Rung-Ching Chen, Hasbi Yasin, Suhartono Suhartono, Youngjo Lee, Bens Pardamean
Abstract
The exposure rate to air pollution in most urban cities is really a major concern because it results to a life-threatening consequence for human health and wellbeing. Furthermore, the accurate estimation and continuous forecasting of pollution levels is a very complicated task. In this paper, one of the space-temporal models, a vector autoregressive (VAR) with neural network (NN) and genetic algorithm (GA) was proposed and enhanced. The VAR could tackle the issue of multivariate time series, NN for nonlinearity, and GA for parameter estimation determination. Therefore, the model could be used to make predictions, such as the information of series and location data. The applied methods were on the pollution data, including NOX, PM2.5, PM10, and SO2 in Taipei, Hsinchu, Taichung, and Kaohsiung. The metaheuristics genetic algorithm was used to enhance the proposed methods during the experiments. In conclusion, the VAR-NN-GA gives a good accuracy when metric evaluation is used. Furthermore, the methods can be used to determine the phenomena of 10 years air pollution in Taiwan.