Detecting Criminal Firms: A Machine Learning Approach

Abstract

In this paper, we develop a machine learning supervised classification algorithm to detect private firms connected to organized crime using financial accounting data. Using Italy as a research setting, we analyze judicial evidences on shareholders and executives charged for Mafia-crimes and build a sample of 2.082 unique private firms connected to Mafia organizations. Leveraging the fundamental differences in accounting numbers between legal and criminally connected companies and exploiting a XGBoost classification algorithm, we are able to detect criminal connections in a hold-out sample of financial statements with an AUC of 74.9% and a notable precision rate of 91.4%. We propose the approach developed as a risk management tools for corporations and a useful support for legal enforcement actions.

Keywords: Financial Accounting, Criminal Firms, Machine Learning, Criminal Scores