Combining rough set and principal component analysis for preprocessing on commercial data stream

In data stream environment, the amount of data is large and the attributes is multidimensional which cause difficulty to data preprocessing. In order to solve these problems, a heuristic method for reduction of attributes (RSPCA) combining rough set theory (RS) with principal component analysis (PCA) is proposed. Firstly, PCA makes dimension reduction and removes noises for the data on sliding window of data stream. After that, decision-making information composed of principal component variables is got. Then, according to the characteristics of principal components that they are relative independence with each other, equivalence class is divided. On this basis, an attribute reduction algorithm of equivalence class of rough sets is proposed which uses significance of principle component variables as heuristic function. Finally, a minimum reduction of the decision attributes is obtained for follow-up mining. Through applying the RSPCA for data preprocessing of customer churn prediction in a business enterprise, the experiment results suggested that the method has superiority on disposing large quantity and multidimensional attribute of data.