Predicting merchant future performance using privacy-safe network-based features.

Area of Science:

• Business and Economics
• Computer Science
• Data Science

Background:

• Small and Medium-sized Enterprises (SMEs) are vital economic contributors, necessitating reliable performance prediction for business financing.
• Current SME performance prediction relies on sensitive internal data, posing privacy risks for merchants.
• Financial institutions require robust methods to assess SME creditworthiness without compromising confidential information.

Purpose of the Study:

• To develop a privacy-preserving approach for predicting SME future performance.
• To leverage credit card transaction data for merchant performance assessment.
• To offer a secure alternative for data sharing between merchants and financial institutions.

Main Methods:

• Constructed a merchant network where customers act as intermediaries between merchants.
• Extracted network structure features for machine learning model input.
• Compared the predictive performance of network-based features against conventional revenue and customer data.

Main Results:

• Machine learning models utilizing network-derived features achieved predictive performance comparable to models using traditional financial metrics.
• The proposed network-based approach significantly enhances data privacy compared to methods relying on direct revenue or customer data.
• Demonstrated the feasibility of using anonymized transaction network structures for SME performance evaluation.

Conclusions:

• The novel merchant network approach provides a privacy-conscious solution for SME performance prediction.
• This method facilitates safer data sharing, enabling informed lending decisions by financial institutions.
• The study addresses critical privacy concerns in financial assessments of SMEs, promoting economic growth.