Rainfall forecasting by utilizing adaptive neuro-fuzzy inference system in Aceh Besar District

Abstract

Forecasting is a common thing to capture events in future based on previous information. However, some classical time-series methods, including moving average (MA), autoregressive integrated moving average (ARIMA), seasonal autoregressive integrated moving average (SARIMA), and simple exponential smoothing (SES), have limitations in predicting nonlinear time-series data. Therefore, this paper aims to utilize the adaptive neuro-fuzzy inference system (ANFIS) model, a combination of the fuzzy inference system (FIS) and neural network architecture to forecast a nonlinear rainfall problem. This model can capture the non-linear data, adaptation capability, and speedy learning capacity. We used the data consisting of temperature (ºC), humidity (%), and wind speed (km/hour) as input variables and rainfall (millimeter) as an output variable at two stations and one rain post in Aceh Besar District, from January 2009 to December 2019. The results demonstrated that ANFIS with generalized Bell (gBell) membership function on epoch 10 can successfully conduct rainfall forecasting in Aceh Besar District with the best-predicted value. The mean absolute percentage error (MAPE) of the prediction at the Meteorology, Climatology, and Geophysics Agency (MCGA) Station or Badan Meteorologi, Klimatologi dan Geofisika (BMKG) Indrapuri is 6.73% for 80% of the training dataset and 20% of the testing dataset.