Febrianto Alqodri, Triyanna Widiyaningtyas, Didik Dwi Prasetya
This study compares Autoregressive Integrated Moving Average (ARIMA), Random Forest, and Bidirectional Long Short-Term Memory (BiLSTM) models for predicting the Dollar Index (DXY) close prices using data sourced from Yahoo Finance, spanning January 1, 2001, to February 10, 2025. The highly volatile, dynamic, and non-linear fluctuations of financial markets underscore the importance of accurate forecasting. Performance is evaluated using MAPE, RMSE, R2, execution time, and memory usage on a test set of 1, 2 1 8 samples. BiLSTM outperforms with a MAPE of 6.914 × 10-5(0.006914%), RMSE of 0.007996, and R2 of 0.9999998, but requires 12.900 seconds and 7.030 MB. Random Forest and ARIMA exhibit comparable accuracy (MAPE approx 0. 0 5 2 9), yet their negative R2 values (-0. 0 0 3 1 and - 0. 0 0 6 1) indicate limited explanatory power. Random Forest is computationally efficient (0.563 seconds, 0.316 MB), while ARIMA demands 20.951 MB and 28.484 seconds. BiLSTM is optimal for high-accuracy DXY forecasting in resource-rich environments, with Random Forest as a faster alternative. These findings enable the public and industry to project future USD prices, mitigate risks, and capitalize on price differences. Moreover, this study serves as a foundation for predictive developers to assess the Dollar Index's influence on currency pairs, such as in forex markets. © 2025 IEEE.
Universitas Negeri Malang, Department of Electrical Engineering and Informatic, Malang, 65145, Indonesia
