<![CDATA[The exchange rate between the Indonesian Rupiah and the United States Dollar (USD) is a critical macroeconomic indicator that shapes international trade, foreign investment, and monetary policy. However, machine learning-based forecasting studies that specifically target the Jakarta Interbank Spot Dollar Rate (JISDOR) as the primary output variable remain limited, leaving a notable gap in the Indonesian financial forecasting literature. This study employs a quantitative time-series approach to address that gap by applying Support Vector Regression (SVR) with a two-stage grid search hyperparameter optimization to JISDOR data spanning January 2023 to January 2025, and benchmarks its performance against an Artificial Neural Network (ANN). Nonlinearity in the data was confirmed through the Teräsvirta test, which justified the use of kernel-based nonlinear models. The optimized SVR with a Radial Basis Function (RBF) kernel achieved an RMSE of 43.49 and a MAPE of 0.22% on the test set, outperforming the best ANN architecture which recorded an RMSE of 49.72 and a MAPE of 0.25%. Neither model showed signs of overfitting, as the errors on the test set were consistently lower than those on the training set. These findings indicate that optimized SVR serves as a reliable and practically applicable forecasting tool for the Rupiah to USD exchange rate. For policymakers, the model offers a quantitative basis for determining the appropriate timing of market interventions, while for financial market participants, it supports more precise hedging decisions and exchange rate risk management.]]>
