Using Machine Learning to Predict Success or Failure in Chapter 13 Bankruptcy Cases

Abstract

Obtaining a chapter 13 bankruptcy discharge is notoriously difficult. Past empirical studies conclude that only one-third of chapter 13 debtors complete their obligations under their plans and obtain a chapter 13 discharge. Many cases end up dismissed, or converted to a case under chapter 7. New data recently made available by the Federal Judicial Center, shows that in recent years only about 39% of chapter 13 filers successfully obtain their chapter 13 discharges. These are low numbers. In this project I examined a public case level database made available in 2017 by the US Federal Judicial Center, based on information collected by the Administrative Office of the United States Courts. The project examines the extent and quality of this data, and the steps needed to use it for advanced statistical analysis and application of machine learning models. This project goes beyond such descriptive statistics. Using machine learning algorithms – so-called artificial intelligence – it describes a model that can predict, using data from the Federal Judicial Center's Integrated Database, whether a debtor will obtain a chapter 13 discharge based only on information provided in the initial petition and summary of schedules. The model is able to predict case results with 70% accuracy overall – and for about 25% of cases can predict results with more than 90% accuracy. When case predictions are cross-referenced against actual case results, the model can assign to specific cases a highly accurate probability of success. The model uses a random forest decision tree algorithm to achieve its results, although nearly similar results were also obtained using a neural network. The model, relevant scripts, and related files and instructions for use are available online through Github at /warrenagin/Ch13Learner.