Rainfall Forecasting Multi Kernel Support Vector Regression Seasonal Autoregressive Integrated Moving Average (MKSVR-SARIMA)

Rainfall variation in the tropics is caused by several factors, such as: geographic, topographical, and orographic. Therefore, the importance of rainfall analysis is needed to know the factors also local characteristics that affect fluctuations in daily rainfall / monthly in each particular area. Rainfall is one element of weather that has a vital role in various sectors in Indonesia. In the agriculture sector, rainfall prediction is used to know the schedule prediction of cropping pattern to optimize food crop production result. In the land, sea and air transport sector, the weather factor that rainfall has a role in the level of safety. In this paper, we used daily rainfall data in Manado, North Sulawesi province in January 2017-December 2017. In short, we combined of SARIMA, and Localized Multi Kernel Support Vector Regression (LMKL SVR) with linear kernel and polynomial kernel reached accuracy model R-2 98.76%. On the one hand, after obtained rainfall prediction, we compared with actual rainfall data in January 2018-February 2018 (59 data). Mainly, Rainfall is difficult to predict even though the model obtained has good accuracy. Still, after validation data forecast and actual data, there is a very far different with RMSE amount 24.43 because the data climate is very dynamic also there are variables that need to be analyzed in building prediction model of rainfall