An empirical study of concept drift detection for the prediction of TAIEX futures

Financial market data is intrinsically dynamic, because it is usually generated in a sequential manner. Such dynamics are usually associated with concept drift, which indicates changes in the underlying data distribution. In this paper, we present our current work that extends from our previous work where we applied data stream mining techniques to the prediction of Taiwan Stock Exchange Capitalization Weighted Stock Index Futures, or TAIEX Futures. In order to analyze the type of concept drift existing in the TAIEX Futures data, we study various methods and propose an ensemble based method. The proposed method uses Drift Detection Method to determine the number of instances given to a sub-classifier that is a component of an ensemble and corresponds to a concept. By observing changes of relative weights of sub-classifiers, we can determine whether a concept occurs repeatedly. Moreover, compared to another ensemble based method, the proposed method achieves higher accuracy without knowing a parameter that is important for another method.