MACHINE LEARNING-BASED CREDIT SCORING FOR MANUFACTURERS AND IMPORTERS

Abstract

This article proposes “a credit scoring system based on artificial intelligence” designed to support producers and importers of goods, compensating for the inefficiency of traditional methods of assessing creditworthiness. Traditional credit scoring often leads to inaccurate risk assessments, financial losses for lenders, and limited access to financing for businesses. The proposed solution combines “reinforcement learning” and “predictive modeling”, using alternative data sources for a more detailed and dynamic analysis of borrowers' behavior. Experimental evaluation using open datasets demonstrates “a 15% reduction in the root-mean-square error”, “an increase in average accuracy by 10%” and consistently high classification rates (accuracy > 80%, completeness > 75%, ROC-AUC > 0.85), exceeding traditional models in accuracy by 15%. The architecture of the system focuses on “data confidentiality, minimizing bias and explainability”, ensuring transparency in decision-making and compliance with financial regulations. The study highlights the potential “of system scalability across industries” and lays the foundation for further development of ethical and adaptive creditworthiness assessment technologies.