Detection of Financial Health Deterioration Using a Naive Bayes Classifier and Bayesian Measures of Surprise

Financial health deterioration may imply financial distress and fraud. Nevertheless, it is computationally intensive to signal financial health deterioration earlier since new financial data are periodically available, and these data may outdate an old detection of financial health deterioration. This study integrates a Naive Bayes classifier with the Bayesian measure of surprise to detect financial health deterioration. This Bayesian measure of surprise denotes the anomaly measure, whereas the Naive Bayes classifier classifies past financial ratio data to create a classification model. Defining the Bayesian measure of surprise employs the relative posterior predictive surprise. For seamless integration with the Naive Bayes classifier’s workflows and pipelines, computing the Bayesian measure of surprise is a post-processing step in implementing the Naive Bayes classifier. Our detection targets are those cases subjected to relative posterior predictive surprises below 0.01. Experiments on actual financial data show that the Bayesian measure of surprise complements a Naive Bayes classifier in detecting assumed anomalies without labeling past normal or anomalous data and calibrating many parameters. However, the Bayesian measure of surprise helps signal assumed anomalies earlier than the p-value does. In conclusion, a Naive Bayes classifier and the Bayesian measure of surprise can be better tools for anomaly detection.