Open Access Open Access  Restricted Access Subscription Access

Adaboost-Based Long Short-Term Memory Ensemble Learning Approach for Financial Time Series Forecasting


Affiliations
1 School of Economics and Management, North China Electric Power University, Beijing, 102206, China
 

A hybrid ensemble learning approach is proposed for financial time series forecasting combining AdaBoost algorithm and long short-term memory (LSTM) network. First, LSTM predictor is trained using the training samples obtained by AdaBoost algorithm. Then, AdaBoost algorithm is applied to obtain the ensemble weights of each LSTM predictor. The forecasting results of all the LSTM predictors are combined using ensemble weights to generate our final results. Four major daily exchange rate datasets and two stock market index datasets are selected for model evaluation and model comparison. The empirical study demonstrates that the proposed AdaBoost-LSTM ensemble learning approach outperform other single forecasting models and other ensemble learning approach in terms of both level forecasting accuracy and directional forecasting accuracy. This suggests that the AdaBoost-LSTM ensemble learning approach is a highly promising for financial time rates forecasting.

Keywords

Adaboost Algorithm, Ensemble Learning, Financial Time Series Forecasting, Long Short-Term Memory Network.
User
Notifications
Font Size


  • Adaboost-Based Long Short-Term Memory Ensemble Learning Approach for Financial Time Series Forecasting

Abstract Views: 887  |  PDF Views: 172

Authors

Yungao Wu
School of Economics and Management, North China Electric Power University, Beijing, 102206, China
Jianwei Gao
School of Economics and Management, North China Electric Power University, Beijing, 102206, China

Abstract


A hybrid ensemble learning approach is proposed for financial time series forecasting combining AdaBoost algorithm and long short-term memory (LSTM) network. First, LSTM predictor is trained using the training samples obtained by AdaBoost algorithm. Then, AdaBoost algorithm is applied to obtain the ensemble weights of each LSTM predictor. The forecasting results of all the LSTM predictors are combined using ensemble weights to generate our final results. Four major daily exchange rate datasets and two stock market index datasets are selected for model evaluation and model comparison. The empirical study demonstrates that the proposed AdaBoost-LSTM ensemble learning approach outperform other single forecasting models and other ensemble learning approach in terms of both level forecasting accuracy and directional forecasting accuracy. This suggests that the AdaBoost-LSTM ensemble learning approach is a highly promising for financial time rates forecasting.

Keywords


Adaboost Algorithm, Ensemble Learning, Financial Time Series Forecasting, Long Short-Term Memory Network.

References





DOI: https://doi.org/10.18520/cs%2Fv115%2Fi1%2F159-165