Due to easily-correlated and multi-index of indicative attributes in churn data on chain retail industry, prediction model based on Support Vector Machine (SVM) was set up. Principal Component Analysis (PCA) can realize dimension reduction and eliminate redundant information, make the sample space for SVM more compact and reasonable. In this paper, PCA was adapted firstly to process 31 dimensional feature vectors of customer churn data, then with the application and verification in real chain retail data set, it was demonstrated that this model based on PCA and SVM has a better performance than the prediction based on SVM only and others.