CHALLENGES AND FUTURE TRENDS OF MACHINE LEARNING IN DIGITAL FINANCE: AN ANALYSIS OF INTERPRETABILITY, REGULATION, AND DATA GOVERNANCE

Abstract

Machine learning (ML) technologies are transforming digital finance through applications in credit assessment, fraud detection, and algorithmic trading. However, their deployment faces three critical challenges: model interpretability, robust data governance, and complex regulatory compliance. This paper analyzes these challenges through a systematic examination of recent literature and regulatory developments. We find that the "black-box" nature of complex models conflicts with transparency requirements mandated by financial regulations such as the EU AI Act and GDPR. Data quality issues, including class imbalance and inconsistency, coupled with privacy concerns, further constrain model reliability. Privacy-preserving approaches, particularly federated learning, offer promising solutions but require wider adoption. We identify that current model governance frameworks lack standardization across institutions and jurisdictions. Our analysis suggests that addressing these challenges requires coordinated efforts across three dimensions: advancing explainable AI (XAI) techniques, establishing unified model governance standards, and implementing privacy-preserving technologies. This study contributes to the understanding of socio-technical barriers in financial ML adoption and provides guidance for practitioners and policymakers.