A Hybrid Recurrent Neural Networks Model Based on Synthesis Features to Forecast the Taiwan Stock Market

Recently, many academy researchers have proposed several forecasting models by technical analysis to forecast stocks, such as (Yamawaki & Tokuoka 2007) [1]. The traditional approach uses a linear time series model for stock forecasting. However, the results would be in doubt when the forecasting problems are nonlinear. Multifeature data from financial statements usually produce high-dimensional data, and therefore, the proposed model utilizes synthesis feature selection for reducing the number of dimensions. The proposed hybrid model utilizes synthesis feature selection to optimize the recurrent network (RNN) for predicting stock price trends. Three refined processes are proposed in the hybrid model for forecasting: (1) select essential technical indicators from popular indicators by a correlation matrix; (2) use stepwise regression and a decision tree to reduce features; and (3) utilize a recurrent neural network (Elman neural network) to build a forecasting model. A six-year period of the Taiwan stock exchange capitalization weighted stock index (TAIEX) is employed as a verification database to evaluate the proposed model under a performance indicator, root mean squared error (RMSE). The results show that the proposed model is superior to the listing models.