The purpose of this study is to predict bankrupt companies by multi-class data mining techniques and to identify the attributes associated with bankruptcy in U.S. publicly listed companies. A corporate bankruptcy has dramatic effects on economy. The 2008 financial crisis can be considered as a noticeable example that led to a high number of bankruptcies. Bernanke (1981) argues that due to substantial social costs from bankruptcy, all participants are interested in preventing it. To avoid occurrence of something someone needs to predict it. Therefore, extensive research to develop accurate prediction models for corporate bankruptcy is justified. Most bankruptcy studies have attempted to develop prediction models using different statistical and/or artificial intelligence methods and techniques. There is a growing literature in financial distress prediction because of its uses in a variety of contexts, like monitoring the solvency of financial institutions, going-concern evaluations by auditors, assessment of corporate loan security, the pricing of credit derivatives, corporate valuations, and portfolio analysis.
The objective of this study is to develop a prediction model that considers various company financial ratios and non-financial variables to explain the bankruptcy in the U.S. publicly listed companies. For bankruptcy prediction and identifying the related attributes, a sample of U.S. companies listed on NYSE, AMEX, and NASDAQ will be selected. The data will be obtained from Compustat, CRSP, and Audit Analytics. The project will use five data-mining techniques to examine the relationship between explanatory variables and bankruptcy and to build a prediction model: Support Vector Machines (SVM), K-Nearest Neighbors (k-NN), Probabilistic Neural Network (PNN), Decision Trees, and Rough Sets.