Hybrid ARIMA-ANFIS for Rainfall Prediction in Indonesia

In Indonesia, rainfall prediction is very important especially for food production. The M3-competition shows that more complicated model not always yield better forecast than simpler one. Conversely, from this competition there is a statement shows that when the various methods are being combined, the accuracy is better than the individual method. This paper proposed hybrid and ensemble model of forecasting method for ten-daily rainfall prediction based on ARIMA (Autoregressive Integrated Moving Average) and ANFIS (Adaptive Neuro Fuzzy Inference System) at six certain area in Indonesia. To find an ensemble forecast from ARIMA and ANFIS models, the averaging and stacking method was implemented. In this study, Triangular, Gaussian, and Gbell function are used as membership function in ANFIS. The best model is measured by the smallest root of mean square errors (RMSE) at testing datasets. The results show that an individual ARIMA method yields more accurate forecast in five rainfall data, whereas ensemble averaging multi model yields better forecast in one rainfall data. In general, these results in line with M3 competition results that more complicated model not always yield better forecast than simpler one.