Research On Multi-class SVM And Its Application In Mutual Funds Evaluation

Support Vector Machine(SVM)is a new tool invented by Vapnik in1990s,which can solve machine learning problems with optimal methods.Great progress has been made both in theory and application,and multi-class SVMs have been a hot research point.The main research we have done are as follows:Firstly,we expounded the theory and algorithm thought,and do contrastive analysis on the performance of available multi-class algorithms through empirical analysis.A new multi-class SVM based on the optimal binary tree to solve the problem of "rejection of classifying""accumulated error" etc.According to the comparison test,the new algorithm perform better in time and accuracy than traditional approaches such as OAA-SVMs、OAO-SVMs、DT-SVMs.Secondly,we elaborate domestic and foreign approaches on investing funds evaluation,and gave the good and bad points of existing investing funds evaluation system.Finally,empirical analysis was done on investing funds evaluation with multi-class SVM. We used financial indicators of the fund as input vector,after feature selection with PCA and choosing proper kernel function and parameters with N-fold cross validation,used two kinds of multi-class SVMs to do evaluation on sample mutual funds.The result showed this model perform well in mutual funds evaluation, and the highest accuracy reached80%.Besides,the new method we offer-DT-SVMs perform better than OAA-SVMs.