PCA-SVM Model Building and Application for Financial Distress Predication of Listed Companies

Abstract

Principal Component Analysis (PCA) is a technique to simplify data collection. Support Vector Machine (SVM) has advantages of suitability of dealing with small sample problem, high dimension, and strong performance of generalization, etc. Therefore, this paper combines the two methods, and proposes a flnancial crisis predication model by integration of principal component analysis into support vector machine. Firstly, complete data pre-processing through extracting principal components of sample set, compress efiectively dimension of sample set, simplify input vectors, and eliminate col-linearity between variables; secondly, train SVM through training set, and on this basis make use of flnancial. data of listed companies to conduct empirical analysis of flnancial predication. Simulation results show that flnancial crisis prediction model built based on PCA-SVM algorithm has superior learning ability and generalization ability, can efiectively reduce the dimensions of sample set, and further improve accuracy of the flnancial prediction, with good feasibility and practical signiflcance.