The SOFM-SVM Model Based On The DPRIF And Its Application

Based on the theory and methods of data pattern recognition, this thesis focused on the disadvantages of the unaided use with unsupervised clustering and supervised classifying method to classify the unlabeled data samples, and furtherly proposed a method of "C2CMA"(from clustering to classification mining application), which integrates clustering with classifying method to solve the problem of pattern recognition. An integrated strategy called "DPRIF"(Data Pattern Recognition Integrated Frame) was also proposed, which is used to discover and explain data pattern when sample set is lack of labeled information. It can also be used to create a classifier which is more stable and more accurate, and can be used to discriminate and forecast new data. Under the guidance of DPRIF integrating strategy, we set up an integreted SOFM-SVM model. Then we analyzed and optimized the model from the aspects of operating mechanism, data interface and function expandedness. The PCA method was introduced to reduce dimensionality and extract features, then to strengthen the clustering explanation; defined a CMI index, to ascertain the most effective or the best clustering number; A new Anti-NO algorithm was proposed to recognize and to filter the suspecious data in the sample; The medium result of SVM model was used to extract the borderline datas between two classified groups. This research compensated for the achievements of data recognition including pattern data, noises data and borderline data. Finally the integrated model was used in the recongnition of listed companies' credit pattern. The results proved that this DPRIF and SOFM-SVM model is effective in solving C2CMA problem. Furthermore, we explored in quantitative research on the issue of stock's credit pattern recognition.