Development of Support Vector Regression Models for Northeast Monsoon Rainfall Prediction

This study has developed nine Support Vector Regression (SVR) models for predicting the Northeast Monsoon Rainfall NEMR) over the peninsular region of India. The Sea Surface Temperature (SST) anomaly was used as an input predictor for the aforesaid prediction task. The performance scores of these SVR models were investigated by changing the epsilon parameter (e=0.1, 0.2,., 0.9) values. The regularization parameter is fixed to a particular value by applying a hit-and-trial method based on the prediction results. Among all the developed models, it is observed that the SVR1 model provides better accuracy with minimum prediction error scores. The goodness of fit measure has also shown the better predictability of the SVR1 model (R2 >0.8) for NEMR prediction in monthly and seasonal time scales. This paper reveals that SST anomaly has the potential to be used as a predictor for NEMR predictability assessment and the developed SVR models with proper parameter tuning may provide optimal prediction outcomes.