Financial Distress Data Mining by Graph Regularized Non-negative Matrix Factorization

• Main Authors: Yu-Cheng Hsieh, 謝侑澄
• Other Authors: Shian-Chang Huang
• Format: Others
• Language: zh-TW
• Published: 2013
• Online Access: http://ndltd.ncl.edu.tw/handle/65634440383243948902

碩士 === 國立彰化師範大學 === 企業管理學系 === 101 === In recent decades, due to the dramatic changes in the global economic, bankruptcy prediction become an important issue for investors and government because the enterprise bankruptcy would incur large losses for investors and increase social costs. Thus, many researches study how to predict whether the company would suffer financial crisis or not. Most of the early warning models were based on financial ratios (i.e. Altman, 1968; Martin, 1977). However, many literatures show that the factors of bankruptcy are not only financial ratios, there are many factors would impact the predict of financial crisis, such as corporate governance, macroeconomic, Audit Opinions, Auditor Changes and audit firm changes. As the reasons, we consider these factors to build the financial distress prediction models. The data are sampled from Taiwan Stock Exchange Corporation (TWSE) from 1999 to 2010, including 111 variables. However, high-dimensional data not only decrease compute speed but also incur curse of dimensionality. For solve this problem, we use Nonnegative Matrix Factorization (NMF) and Graph Regularized Non-negative Matrix Factorization (GNMF) to reduce dimensions, and construct financial distress prediction models by logistic regression, neural network (NN), support vector machine (SVM) and ensemble algorithms-bagging.