Forecasting USD to IDR Exchange Rates Using Prophet Time-Series Model

Abstract

This study evaluates the effectiveness of the Prophet time-series model in forecasting USD to IDR exchange rates using a historical dataset of 2812 daily records, including opening and closing prices, highs, lows, and percentage changes. Data preprocessing steps, such as handling missing values and standardizing numeric fields, were performed to ensure data quality. Prophet, developed by Facebook, was chosen for its capability to model seasonality, irregular patterns, and external regressors, outperforming traditional models like ARIMA. The model's performance was validated using error metrics, including Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and Mean Absolute Percentage Error (MAPE), demonstrating its predictive accuracy. Comparative analysis with ARIMA confirmed Prophet’s superior ability in capturing complex patterns in volatile financial data. The inclusion of external factors such as inflation rates and global economic indicators further improved the forecast accuracy. The results provide valuable insights for policymakers, investors, and financial analysts, supporting more informed decision-making and risk management strategies. This research highlights the importance of proper data preprocessing and advanced forecasting techniques for improving currency prediction accuracy, especially in emerging markets like Indonesia. Future research could explore hybrid models combining Prophet with machine learning techniques for enhanced forecasting capabilities.