Prediction of the Exchange Rate of the Russian Ruble (RUB) against the United States Dollar (USD): Performance Comparison of LSTM and CNN Models

Abstract

This research aims to compare the effectiveness of Long Short-Term Memory (LSTM) and Convolutional Neural Network (CNN) models in predicting the exchange rate of the Russian Rouble (RUB) against the United States Dollar (USD). Currency exchange rates have complex time series characteristics with high volatility, especially for an economy like Russia that is affected by various geopolitical and economic factors. Both models were trained using historical USDRUB exchange rate data and evaluated based on Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), coefficient of determination (R²), and prediction accuracy metrics. The results showed that the LSTM model outperformed CNN on all evaluation metrics with RMSE of 4.42 (versus 4.99 for CNN), MAE of 1.67 (versus 2.00 for CNN), MAPE of 1.76% (versus 2.12% for CNN), and R² of 0.8775 (versus 0.8079 for CNN) on the test data. These findings indicate that the LSTM's ability to model long-term dependencies provides a significant advantage in predicting currency exchange rates compared to convolution-based approaches. This research provides important insights for monetary policy makers, financial market analysts, and international business people who depend on accurate exchange rate predictions for strategic decision making.