Comparative Study of Kernel Function for Support Vector Machine on Financial Dataset

摘要

Due to the increasing number of business failures effect from economic crisis, it is challenging to develop a financial distress prediction model. The prediction model is the early warning system that has any advantages for companies, consumer, creditors investors and the economy of country in general. We develop SVM Model with different kernel function such as linear, polynomial and radial basis function. We purposed tuning method with 10-fold cross validation to find the best pair of parameters for each kernel function. The result shows that SVM model using radial basis kernel with optimal parameter C = 5 and y = 1 is obtain appropriate accuracy, the AUC value is 0.72.