A genetic algorithm based feature selection approach for rainfall forecasting in sugarcane areas

Rainfall is a vital phenomenon that contributes in the success of sugar industry season. The ability to determine the amount of precipitation in sugarcane areas enhances the profitability of the season. Different types of climate indices and attributes are usually applied to model rainfall forecasting systems. In this paper, we present a novel genetic algorithm based feature selection approach to determine which climate indices and attributes are most significant for rainfall forecasting in sugarcane areas. The most significant features are features that return the highest accuracy for rainfall forecasting through artificial neural networks. The approach is evaluated on realworld data that contain different weather forecasting features. A set that contains maximum temperature values and Southern Oscillation Index (SOI) has proven to be the best combination among the other models with a Root Mean Square Error (RMSE) of 0.027 in November. An Average RMSE of 0.0638 for the genetic algorithm based forecasts was recorded. The proposed model was compared to other models and the proposed model revealed higher accuracy in forecasting monthly rainfall.