Hybrid CNN-LSTM Architecture for Foreign Exchange Rate Prediction Using Technical and Fundamental Analysis Features

Abstract

Accurate prediction of foreign exchange (Forex) rates is a challenging task due to the non-stationary, chaotic, and highly non-linear nature of currency markets. Traditional statistical models often fail to capture the complex temporal dependencies inherent in financial time series, while standalone deep learning approaches may lack the spatial feature-extraction capability required for robust forecasting. This paper proposes a novel hybrid deep learning architecture that combines Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) networks for one-step-ahead Forex rate prediction. The CNN layers automatically extract local spatial patterns from multi-variate input sequences, whereas the stacked LSTM layers model long-range temporal dependencies across time. A comprehensive feature set derived from both technical indicators (RSI, MACD, Bollinger Bands, SMA, ATR) and macroeconomic fundamental variables (interest-rate differentials, CPI ratios, GDP growth differentials) is jointly encoded within a unified sliding-window framework. The proposed model is evaluated on publicly available daily EUR/USD, GBP/USD, and USD/JPY exchange-rate data covering the period 2010-2024. Experimental results demonstrate that the CNN-LSTM hybrid significantly outperforms baseline models including standalone CNN, LSTM, GRU, and ARIMA in terms of Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and Mean Absolute Percentage Error (MAPE). The model achieves an RMSE of 0.0023 on EUR/USD, representing an improvement of 18.4 % over the best individual benchmark. These findings confirm that combining convolutional spatial extraction with recurrent temporal modelling, augmented by macroeconomic fundamentals, yields a practically effective and computationally efficient Forex forecasting system.