Data preprocessing for stock market forecasting using random subspace classifier network

Financial forecasting is a challenging problem which attracts researchers from different fields. Since many authors suggest new methods to predict the market, it´s reasonable to suppose that the reliable solution is hard to find. Moreover, discovered regularity in the market behaviour should not exist for a long period of time, because market participants will try to exploit that opportunity. In order to predict the future with some degree of confidence it´s better to make sure the forecasting method provides expected results for different time frames. This article analyses the application of random subspace classifier for predicting the next day stock price return. Different data preprocessing approaches, particularly for stock price normalization, are suggested. Forecasting performance is tested for different time periods. Besides, the ability of the network to predict the price change is considered within the test set.