Rainfall is related to a number of factors that are interdependent and influenced by dynamic global time, region and climate factors. Determination of relevant predictors is important for the efficiency of the rainfall estimator model. Although some climate modeling studies in one region/country have high accuracy, this model is not necessarily suitable for other regions. Determination of predictor variables by considering spatio-temporal factors and local / global features results in a very large number of inputs. Feature selection produces minimal input so that it gets relevant predictor variables and minimizes variable redundancy. Recurrent Neural Networks is one of the artificial neural networks that can be used to predict time series data. This study aims to predict rainfall by combining the SVM classification method and the RNN method. Tests on the Perak 1 daily and monthly weather data (WMO ID: 96933) and Perak 2 Station daily and monthly data(WMO ID: 96937), showed high accuracy results with an R2 are 92.1%; 94.1%; 90.9% and 89.6%.