Prediction and estimation of atmospheric pollutant levels by soft computing approach