Stock investment decision support using an ensemble of L-GEM based on RBFNN diverse trained from different years

Many researches attempt to find out regularities of the stock market so as to gain higher profit within stock investment activities. In fact, the price changes over time, but the data from previous period reflects the future trend in some extent. Thus, a new method is proposed to investigate how to use the historical data to make correct investment decision in this paper. Since the technical indicators are efficient tools on stock prediction, we use technical indicators of every trading day in current year as the input variables to train a RBFNN based on L-GEM and make prediction of trading actions (buy, sell or hold) for days in the next year. Then, a Multiple Classifier System (MCS) is built to make the final prediction from all base RBFNNs which trained by data in different years. Experimental results show that the new information from past stock data is predictable. Data in different years causes different effects on the future stock market. Accumulating all information from base RBFNNs brings a considerable profit in our experiments.