Application And Research Of SVM On Finance Early Warning System

Through adoption of structure risk minimization principle, Support Vector Machine (SVM) which is one new machine learning method based on the solid theory foundation of Statistical Learning Theory solves the little-sample statistic learning problem, and is becoming the new research focus after neural network with such advantages as using kernel function to avoid local minimal point, sparse nature of solutions, limit used to control capacity or the number of support vectors.The thesis describes and explores the theory foundation of SVM, and makes a summarization for present studies, a comparison between main improved algorithms. On this basis and an analysis on the main problem in SVM, especially in the selection and turning of SVM parameters in specific method. On this basis, combination of Principal Component Analysis (PCA) and SVM, a Finance Early Warning Model applicable to different industry and little-sample space is provided which decides the enterprise status based on traditional finance guideline and through the main element analysis to simplify the input.Abstract the inter correlation through SVM between the excellent enterprise rates, under such data condition as small sample, high dimension, big noise and nonlinear relationship, the thesis analyzes the enterprise finance data and gets the accurate prognostication for enterprise status assessment. This method overcomes the localization of linearity distinguish in old models and needs less samples to train the net than BP neural network.