<![CDATA[Exchange rate volatility of the Indonesian Rupiah against the US Dollar has increased due to global uncertainty. This study addresses the limitation of prior research that predominantly relies on single linear or nonlinear models in emerging markets by developing a Hybrid ARIMA SVR approach, thereby enhancing exchange rate predictability to support macroeconomic stability. This study contributing to the advancement of quantitative forecasting methods aligned with SDG 8 and SDG 16 through enhanced financial predictability. This research uses a univariate time-series dataset of weekly Rupiah US Dollar exchange rates obtained from Bank Indonesia, comprising 150 observations from March 2023 to January 2026. Novelty from this research is ARIMA model selected to capture linear temporal dependencies, while SVR is employed to model nonlinear patterns in residuals justifying the hybrid approach as a complementary integration of statistical and machine learning methods. Data preprocessing includes Box-Cox transformation and second order differencing to ensure stationarity, followed by diagnostic tests (Ljung Box, Kolmogorov Smirnov, and ARCH LM). SVR parameters are optimized using grid search to ensure robust model performance. The analysis included visualization, Box–Cox transformation (λ = −1), and second-order differencing to achieve stationarity. Diagnostic tests (Ljung Box, Kolmogorov Smirnov, ARCH LM) confirmed that ARIMA (3,2,0) met model assumptions. ARIMA residuals were subsequently model using SVR, with parameters optimized through grid search, forming the Hybrid ARIMA–SVR model. Results show that the Hybrid ARIMA SVR model outperformed the standalone ARIMA, achieving a lower MAPE. The best performance (MAPE = 0.56%) was obtained using the Radial kernel with ε = 0.2, C = 23, and γ = 28. These findings indicate that integrating linear and nonlinear models improves forecasting accuracy.]]>
