Double Exponential-Smoothing Neural Network for Foreign Exchange Rate Forecasting

Abstract

One of the most used method for forecasting is Artificial Neural Network (ANN). The success of ANN to solve the problem depends on the input data. Improving data quality can be done by smoothing the input data. In this study, smoothing data will be done using Exponential Smoothing (ES) approach. We use exchange rate of Indonesia Rupiah (IDR) against US Dollar (USD) from January 2016 to December 2017 for the data research. This research the forecasting using ANN with smoothing process in the data input using Double Exponential Smoothing (DES) will compared with the forecasting using ANN with original data input and forecasting using ANN with smoothing process in the data input using Single Exponential Smoothing (SES) as a model. The model’s performance will have measured using error value and execution time. This research concludes that Double Exponential Smoothing (DES) method can improve the performance of ANN on IDR/USD exchange rate forecasting, it produces 0.530% of MAPE values and takes 561s for time execution, and also, we conclude that DES is better than SES to improve ANN performance for exchange rate forecasting.