Adaboost-based Integration Framework Coupled Two-stage Feature Extraction with Deep Learning for Multivariate Exchange Rate Prediction

• Published in: Neural processing letters Vol. 53; no. 6; pp. 4613 - 4637
• Main Authors: Wang, Jujie, Chen, Yu
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.12.2021; Springer Nature B.V
• Subjects: Accuracy; Algorithms; Artificial Intelligence; Clustering; Complex Systems; Computational Intelligence; Computer Science; Crude oil prices; Deep learning; Econometrics; Feature extraction; Feature selection; Forecasting; Foreign exchange markets; Foreign exchange rates; Interdisciplinary subjects; Machine learning; Model accuracy; Multivariate analysis; Neural networks; Recurrent neural networks; Securities markets; Self organizing maps; Time series; Variables; Deep neural network; Ensemble prediction; Multiple influencing factors; Two-stage feature extraction
• ISSN: 1370-4621; 1573-773X
• DOI: 10.1007/s11063-021-10616-5

The foreign exchange market plays an important role in the financial field. Accurately predicting the exchange rate appears to be difficult on account of the characteristics of time variability and randomness. This study proposes an Adaboost-based reinforcement ensemble learning framework, which combines two-stage feature extraction with deep learning models to perform multivariate exchange rate prediction. Considering the impact of data information hidden in other financial markets on the foreign exchange market, multiple exogenous variables are introduced as input factors of the proposed model. Auto-encoder and Self-organizing map, as the main two-stage feature extraction models, have their advantages in simplifying model input and clustering similar feature data respectively. Feature extraction paves the way for the subsequent establishment of deep recurrent neural network (DRNN) for prediction, which improves the robustness of the model while improving the prediction accuracy. Finally, the Adaboost algorithm is utilized to integrate the DRNN prediction results. The empirical results reveal that the proposed model has higher accuracy in exchange rate prediction. The prediction effect of the model is significantly better than comparable models and it is a promising way of forecasting exchange rates.