Using forensic analytics and machine learning to detect bribe payments in regime-switching environments: Evidence...

Area of Science:

• Forensic Analytics
• Financial Data Science
• Behavioral Economics

Background:

• Benford's Law is a tool for detecting anomalies in numerical data.
• India's 2016 demonetization significantly altered the distribution of legal tender notes.
• Retail transaction data can reveal patterns indicative of financial misconduct.

Purpose of the Study:

• To assess the efficacy of machine learning models and Benford's Law in detecting bribe payments.
• To evaluate the impact of regime shifts, specifically demonetization, on the performance of these models.
• To understand the role of institutional settings in financial data analytics.

Main Methods:

• Utilized random forest and XGBoost models trained on transaction data and bribe payment datasets.
• Employed empirical measures derived from Benford's Law for model guidance.
• Evaluated model performance using pre- and post-2016 Indian demonetization data.

Main Results:

• Models achieved high F1 scores (around 90%) when trained and tested within the same pre- or post-demonetization period.
• Model performance decreased sharply (below 10% F1 score) when trained in one period and tested in another.
• Benford's Law criteria and the models showed meaningful information for bribe detection under stable conditions.

Conclusions:

• Machine learning models combined with Benford's Law are effective for detecting financial irregularities like bribes.
• The effectiveness of financial data analytics is highly sensitive to institutional and environmental regime shifts.
• Future applications require careful consideration of historical context and potential changes in data-generating processes.