SELECTIVE PRESENTATION LEARNING FOR NEURAL-NETWORK FORECASTING OF STOCK MARKETS

This paper proposes a selective presentation learning technique for im proving the learnability and predictability of large changes by back-p ropagation neural network. Daily stock prices are predicted as a compl icated real-world problem, taking nonnumerical factors such as politic al and international events into account. Training data corresponding to large changes of prediction-target time series ave presented more o ften, and network learning is stopped at the point that has the maxima l profit. When this technique is applied to daily stock-price predicti on, the prediction error on large-change data was reduced by 11%, and the network's ability to make profits through experimental stock-tradi ng was improved by 67% to 81%, in comparison with results obtained usi ng conventional learning techniques.