Analisis Kinerja Model ARIMA dan LSTM dalam Memprediksi Jakarta Interbank Spot Dollar Rate (JISDOR)

Abstract

JISDOR exchange rate data is a collection of time series data with certain patterns, such as trend, seasonal, horizontal, and cyclical. According to Cowpertwait and Metcalfe (2009), one of the time series forecasting techniques is based on statistical mathematical models. As technology develops, forecasting methods are becoming more sophisticated and diverse. This research compares two main methods, namely AutoRegressive Integrated Moving Average (ARIMA) and Long Short-Term Memory (LSTM). ARIMA, a classic method of time series analysis, is used to forecast trends and seasonal patterns. LSTM, a subset of artificial neural networks, promises a solution to long-term dependencies and complex patterns in data. In this study, the stationarity of the data was checked, which showed that the data was not stationary, so differencing was performed. Based on the ACF and PACF plots, the parameter values of p,d,q are (2, 1, 0). Both ARIMA and LSTM models were tested to get the best model based on RMSE and MAPE. The ARIMA model has RMSE 82.537 and MAPE 0.4193, while the LSTM has RMSE 75.807 and MAPE 0.4152. These results show that LSTM is better than ARIMA because the RMSE and MAPE values are lower. Therefore, JISDOR exchange rate forecasting is carried out using the LSTM model.