Fusion of multiple indicators with ensemble incremental learning techniques for stock price forecasting

Abstract

Predicting stock market index is very challenging as financial time series shows highly non-linear and non-stationary patterns. In this paper, an ensemble incremental learning model is presented for stock price forecasting, which is composed of two decomposition methods: discrete wavelet transform (DWT) and empirical mode decomposition (EMD), as well as two learning models: random vector functional link network (RVFL) and support vector regression (SVR). Firstly, DWT and EMD are sequentially combined to decompose the historical stock price time series, followed by RVFL models to analyze the obtained sub-signals and generate predictions. Moreover, ten stock market indicators are used to improve the performance of the ensemble model. Last but not least, incremental learning with RVFL also benefits the performance significantly. To evaluate the proposed DWT-EMD-RVFL-SVR model, stock price forecasting for five power related companies are conducted to compare with seven benchmark methods and two recently published works.