Date of Award


Document Type


Degree Name

Doctor of Philosophy (PhD)

First Advisor

Barbara A. Apostolou


This study examines the usefulness of an artificial intelligence method, case-based reasoning (CBR), in predicting corporate bankruptcy. Based on prior research, CBR is believed to be a viable method of predicting bankruptcy. Hypotheses are developed to test the usefulness of a CBR system and to compare the accuracy of such a system to the model considered to be the benchmark model in bankruptcy prediction, Ohlson's (1980) nine-factor logistic regression (logit) model. Sample data consisting of manufacturing and industrial firms is drawn from the Compustat database in a 20:1 ratio of nonbankrupt to bankrupt firms, consistent with Ohlson's (1980) proportions. Three CBR models representing one, two, and three years before bankruptcy are designed and developed using a CBR development tool, ReMind. Cross-validation is done using a 10% in-period holdout sample as well as a holdout sample of firms from outside the period from which the model is constructed. Three logit models based on Ohlson (1980) representing one, two, and three years before bankruptcy are constructed. The usefulness of the CBR system is determined by examination of type I and type II error rates. Chi-square statistics are used to compare the predictive accuracy of the three CBR models with the three logit models. The results indicate that the CBR method using ReMind is not useful in predicting corporate bankruptcy. It is believed that the small sample of bankrupt firms (relative to the sample size of nonbankrupt firms) contributes to the failure of these CBR models to accurately predict bankruptcy. Compared with two other studies that also use ReMind as development tools, there is evidence that the algorithm in ReMind does not accommodate small sample sizes. The results also indicate that CBR is not more accurate than the Ohlson (1980) logit model. Ohlson's (1980) logit models attain a much higher accuracy rate than the CBR models and appear to be more stable over time than the CBR models.