Fraud/Uncollectible Debt Detection Using a Bayesian Network Based Learning System: A Rare Binary Outcome with Mixed Data Structures

TL;DR

This paper presents a Bayesian network-based system for detecting rare fraud and uncollectible debts in telecommunications, effectively handling mixed data types and outperforming traditional methods.

Contribution

It introduces a Bayesian network approach tailored for rare binary outcomes with mixed data, demonstrating improved detection performance over existing techniques.

Findings

Bayesian networks effectively predict rare fraud events.

The model handles mixed categorical and continuous data.

Performance surpasses linear and non-linear discriminant analysis, and decision trees.

Abstract

The fraud/uncollectible debt problem in the telecommunications industry presents two technical challenges: the detection and the treatment of the account given the detection. In this paper, we focus on the first problem of detection using Bayesian network models, and we briefly discuss the application of a normative expert system for the treatment at the end. We apply Bayesian network models to the problem of fraud/uncollectible debt detection for telecommunication services. In addition to being quite successful at predicting rare event outcomes, it is able to handle a mixture of categorical and continuous data. We present a performance comparison using linear and non-linear discriminant analysis, classification and regression trees, and Bayesian network models