THE APPLICATION OF THE ARTIFICIAL NEURAL NETWORK (ANN) METHOD FOR FORECASTING THE SOUTHERN OSCILATION INDEX (SOI)

Abstract

Indonesia's seasons are influenced by global phenomena such as ENSO. This phenomenon affects rainfall intensity in Indonesia through its two main phases: El Nino and La Nina. One method to detect these events is by analyzing the Southern Oscillation Index (SOI). A highly accurate SOI forecasting model is critical for both short-term and long-term development planning, particularly in anticipating future extreme seasons. One of the methods used for forecasting is the Artificial Neural Network (ANN). This study aims to develop an ANN model capable of predicting the SOI index. Based on forecasting using training data, the optimal model architecture identified is 12-7-1, which achieved the smallest MSE value of 0.0095 and a MAPE of 17.6851. With an error rate below 20%, the 12-7-1 architecture demonstrates strong forecasting capabilities. The study forecasts the SOI index for the next 12 months, indicating a trend from negative values at the beginning of the year to more positive values toward the year's end.