SVR: Feature Selection Based Support Vector Regression for Humidity Data Prediction

Weather data like humidity is a key indicator in meteorology which is influenced by multiple climatic factors. Higher humidity levels become a threat to human health, cause storms and also lead to increase in rainfall, temperature and air borne pollutants. Therefore, prediction of humidity can be helpful in taking necessary action plans for protection of both human life and environment. Weather data can be predicted by multiple regression methods using explanatory variables. However, the effectiveness and prediction accuracy of the multiple regression methods are affected by the irrelevant or non-informative features and the uncertainties lie with the predictions. Feature selection (FS) methods can improve the effectiveness of prediction models and also reduce the uncertainties of predictions by reducing the dimensionality of the dataset by selecting more explanatory variables. On this basis, we have proposed feature selection-based support vector regression (SVR). The proposed hybrid model incorporates Random Forest (RF) model-based FS method in SVR to predict the future humidity data. Performance of RF model-based FS method is compared with two other feature selection methods like correlation-based method and recursive feature elimination (RFE) method. The prediction accuracy of the SVR model is compared with two other multiple regression models: artificial neural network (ANN) regression and decision tree regression (DTR).