Peramalan Data Cuaca Ekstrim Indonesia Menggunakan Model ARIMA dan Recurrent Neural Network

Extreme weather modeling is a challenge for experts in Indonesia and the world. prediction complex problem because chances of it happening are very small, so developed models often have low level accuracy. The purpose this research to combine classic model, Autoregressive Integrated Moving Average (ARIMA), recurrent neural network (RNN) model using Adam SGD estimation (RNN-Adam RNN-SGD) with reLU, tanh, sigmoid gaussian activation functions. In addition, ARIMA-RNN mix was also demonstrated study. These applied monthly period extreme data obtained from Meteorology, Climatology Geophysics Agency (BMKG) West Sulawesi Province which converted into training test data. RMSE value used see accuracy both Based on results, best Indonesia’s ARIMA-RNN-Adam reLU function based At n = 50, smallest MSE values third 0.23212 function, then ARIMA-RNN-SGD 0.25432 same while ARIMA 0.3270. n=100 can be seen that three equal 0.25149 0.25256 0.2644.