Air Pollution Prediction Using an Artificial Neural Network Trained by Chaotic Gravitational Search Algorithm: A Comparative Study

Prediction of urban air pollution is an important subject in environmental studies. However, the required data for prediction is not available for every interested location. So, different models have been proposed for air pollution prediction. The feature selection (among 20 features given in Meteorology Organization data) was performed by binary gravitational search algorithm (BGSA) in this study and 10 features were selected. An artificial neural network (ANN) was used in this study for air pollution prediction. This ANN was trained using chaotic gravitational search algorithm (CGSA). In a comparative study, the performance evaluation of this neural predictor was performed when other methods were also used for ANN training. These methods were error back propagation, standard GSA, artificial bee colony, particle swarm optimization, and hybrid of genetic algorithm and simulated annealing. Experimental results showed the superior performance of the proposed BGSA-CGSA method used for feature selection and ANN training.