ECMWF (European Centere of Medium-Range Weather Forecasts) produces weather data every six hours. In the case of ECMWF 1.125 degree weather data, the northern hemisphere is divided into 320×161 grids and each grid has 254 weather features. Since we are aim to forecast heavy rain in the Korea Peninsula, we need only 10×10 grids around the Korean Peninsula. However, the number of inputs to the forecasting system will be 100 dimensions (10×10) even if we consider only one weather feature. If we consider 3 features, it is 300 dimensions (10×10×3). Therefore, as more features are combined, the size of the data is increased and it causes the computational cost high. In order to reduce the size of inputs to the forecasting system, we apply genetic algorithms for the feature selection in this paper. As a result, it has been found out that it is possible to assort with a higher accuracy rate with a smaller data set.