Efficient air pollutants prediction using ANFIS trained by Modified PSO algorithm

In the modern day world air pollution has been a major concern for various environmental diseases. High traffic density, electricity production, and expanding commercial and industrial activities have increased air pollution in an unpredictable manner. Thus, the forecast of air pollution can be used as an advisory to establish strategies and corrective measures particularly in case of higher air pollution levels. In view of this, in this work we investigate an application of Modified Particle Swarm Optimization (MPSO) to train ANFIS (Adaptive Neuro Fuzzy Inference System) for the efficient prediction of two major air pollutants namely, Sulphur dioxide (SO₂) and Ozone (O₃) in New Delhi. The results obtained are then compared with the traditional gradient based method normally used for training ANFIS. The comparison between the two was based on three performance indices, namely MSE (Mean Squared Error), RMSE (Root Mean Squared Error) and MAD (Mean Absolute Deviation). For SO₂ prediction MSE, RMSE and MAD of 0.110, 0.331 and 0.932 were obtained using the proposed method against 0.117, 0.342 and 1.03 respectively using the traditional method. Similarly for O₃ prediction, MSE, RMSE and MAD of 0.837, 0.700 and 2.49 were obtained against 0.878, 0.812 and 2.93 respectively. These results clearly indicate that ANFIS trained using MPSO is far better at generalizing, achieving higher accuracy for the prediction of SO₂ and O₃ air pollutants.