Can we detect and predict fraud with machine learning?

Abstract

This project aims to construct a machine learning model that is able to detect and predict financial fraud. Based on 3,095 U.S. fraudulent firms, 61 features and 24 years of analysis, this neural network will be able to detect and predict a large part of frauds. Following research and analysis, financial fraud was concluded to have a huge impact, not only for the companies in terms of costs and penalties, but also for investors and for the market as a whole. Many of the fraudulent companies disappear after the discovery of fraud and are unable to recover. The consequences this reflects on the employees, investors and the market are incalculable. Firms undergoing an initial public offering tend to have more incentives to commit fraud, since they are at a critical stage of their life cycle and want to attract as many investors as possible. This project will address these companies separately as well due to their interesting characteristics and evident incentives to commit fraud. This model can be used by investors, by banks in their credit risk models, by venture capitalists and last but not least, institutions like the Securities and Exchange Committee (SEC) that regulate the market.

Date of Award	1 Feb 2021
Original language	English
Awarding Institution	Universidade Católica Portuguesa
Supervisor	Dan Tran (Supervisor)