Using Support Vector Machine and Sequential Pattern Mining to Construct Financial Prediction Model

Prediction models provide investors preliminary information before bankruptcy. Prediction models based on classification technique distinguish a listed company between healthiness and bankruptcy in the most literature, but little attention has been paid to do the further discussion on the sequential analysis of classifications. To supplement this insufficiency, a mixture model of Support Vector Machine (SVM) and Binary Sequential Analysis (BSA) is presented. The BSA mines the predicting pattern from the SVM classification signals to predict next outcome of the company. The mixture modes can not only provide a company the contemporaneous classification but also the next prediction of failure status. Our experimental results of Taiwan stock market reported that the accuracy of BSA prediction is very close to the correctness of SVM classification, or the difference is less than 2%.