Portfolio

Machine Learning for Tax Fraud Detection

Approach:

The project utilized Benford's Law, a statistical principle (first digit rule), to analyze invoice data for unusual number distributions. By training the model on a dataset of fraudulent companies, we created a tool to detect suspicious patterns in invoice numbers and financial reports. We applied machine learning techniques to train the model, focusing on identifying potential tax evasion through patterns in invoice data.

Results:

Successfully identified 420 potentially fraudulent companies.

Achieved an accuracy rate of nearly 90%.

Demonstrated how machine learning can support tax compliance and revenue recovery efforts.

This project highlighted the impact of data science in detecting fraud and contributed to enhancing the effectiveness of tax audits.

About São Paulo State Treasury Department

The São Paulo State Department of Treasury (SEFAZ-SP) oversees US$ 55 billion in annual taxpayer revenue, making it the top ICMS (comparable to the GST) revenue collector among Brazilian states.
With around 8,000 employees, SEFAZ-SP combines modern public management practices with advanced technologies to provide high-quality services both in-person and online.
Its operations span the entire state, with 18 regional tax units, numerous tax posts, and service centers in all 645 municipalities.