Financial Distress Prediction Model Based On Kernel-modified Support Vector Machine

The application of Support Vector Maching, a powerful tool in data miningarea, has been a popular issue for a long time since1990. It is widely appliedespecially in the field of text recognition, face identification, handwritingrecognition, gene classification and time series prediction. However, because thismethodology was just proposed two decades ago, it is still in exploration phase andrequires further research. As a matter of fact, the selection of model and its kerneland parameters is according to prior experience, which means the absence oftheoretical guidance and unguaranteed classification precision. Therefore, the workto perfect the theories, as well as to modify the SVM method is very valuable.The effort in this research is put on the application of kernel-modified SVMalgorithm on the financial crisis prediction field. Note that the selection of kernelshould depend on the data distribution, the idea of this research is to adopt adata-dependent kernel modification algorithm in the modelling. Firstly, the reviewsection illustrates the research background, previous research and significant resultsabout financial crisis prediction based on SVM. Subsequently, focusing on kerneltheories, the key knowledge of Statistical Learning Theory, kernel-relatedconceptions and model parameters optimization is introduced. Next, a kernelmodification algorithm is discussed theoretically, with discussion about expansionamong other adoptable kernels.As the key part of our study, the financial crisis prediction model is built andexperiment is conducted in the last section.165samples of Chinese listedcompanies are collected. By using the Matlab software, the traditional SVM model,PCA-SVM model and proposed SVM model with kernel modification algorithm arecompared according to their performance criteria. The result indicates that theproposed model is a good alternative as it has good performance with respect todecreasing the number of support vectors and improving classification precision. Atlast, we conclude our study by proposing the entire kernel-modified SVM financialcrisis prediction model.