비모수 기법을 이용한 기업 부도예측

Abstract

Diagnosing the firms at risk for bankruptcy is important in financial distress analysis in order to prepare a way of hedging against financial risks from them. There may exist some pre-alarm signals presenting a financial crisis when a firm faces a default risk. Early studies on corporate bankruptcy prediction by detecting the pre-alarm signals include parametric statistical techniques and nonparametric approaches such as artificial intelligence (AI). As one of nonparametric techniques, a support vector machine (SVM) has shown a potential in predicting corporate bankruptcy. We propose a hybrid method that combines data depths and nonlinear SVM to corporate bankruptcy prediction. We employed the data depth functions to condense the multivariate financial data with nonlinear and non-normal characteristics in an one-dimensional space. The SVM method was introduced to classify the data points on depth vs. depth ($DD$) plot. Based on data set which records failed and non-failed manufacturing firms in Korea for 10 years, the empirical results demonstrated that the proposed method offers the highest level of accuracies in bankruptcy prediction. The proposed method is expected to support a guide of corporate investment for investors and other interested parties.

In addition, this thesis, we focus on feature selection and provide an overview of the existing methods that are available for handling several different classes of problems. Additionally, we consider the most important application domains and review comparative studies on feature selection therein, in order to investigate which methods perform best for specific tasks.

Feature selection techniques show that more information is not always good in machine learning applications. We can apply different feature selection methods for the data at hand and with baseline classification performance values we can select a final feature selection algorithm. For the application at hand, a feature selection methods can be selected based on the following considerations: simplicity, stability, number of reduced features, classification accuracy, storage and computational requirements. Overall applying feature selection will always provide benefits such as providing insight into the data, better classifier model, enhance generalization and identification of irrelevant variables. In this thesis used 6 different feature selection method to extract main features and predict possibility of bankruptcy by logistic and ANN method. Also this thesis calculate and compare the accuracy, Type 1 error and type 2 error of each distribution of bankruptcy.