Using Ensemble Learning to Build a Financial Distress System

• Main Authors: En-Chuan Liu, 劉恩銓
• Other Authors: Yi-Chuan Lu
• Format: Others
• Language: zh-TW
• Online Access: http://ndltd.ncl.edu.tw/handle/13839500332061705246

碩士 === 元智大學 === 資訊管理學系 === 103 === This study utilizes Ensemble learning methodology to predict the company financial distress. Total of 93 independent variables are collected from four categories: financial variables, corporate governance variables, stock price variables, and changes of corporate governance variables. Features selection is done by applying two-way step-wise regression algorithm. One-year, two-year, and three-year data before the financial distress were happened are collected for training and testing. The bucket of models (Ensemble) are Neural Network, Support Vector Machine, Logistic Regression, Na#westeur048#ve Bayes, and Decision Tree. The accuracy of Ensemble learning, based on one-year, two-year, and three-year data before the financial distress happened, are 86%、82%、72% respectively, and the sensitivity rates are 84%、84%、74% respectively.