Penerapan Jaringan Saraf Tiruan dengan Metode Backpropagation untuk Peramalan Kurs Dolar Amerika terhadap Rupiah

Abstract

The highly fluctuating exchange rate of the United States Dollar against the Indonesian Rupiah creates significant economic uncertainty, necessitating accurate forecasting tools for proactive policy measures. This research aims to determine the optimal configuration of an Artificial Neural Network model, evaluate its accuracy in model development, and present the forecast results for the exchange rate. The data analysis utilizes a model trained with the backpropagation algorithm and Adam optimization, the model's performance is subsequently evaluated using Mean Absolute Percentage Error. The findings indicate that the optimal model configuration is a 12-5-1 architecture, trained for 4000 epochs with an 80:20 data split for training and testing. This model is highly accurate, demonstrating a Mean Absolute Percentage Error value of 1.8764%. The forecast projects the exchange rate to be within the range of IDR 15,614–IDR 15,850 during the period of January–June 2025.