Improving Credit Decision Through Machine Learning and Alternative Data: Evidence from NBFIS

Non-banking financial institutions (NBFIs) often struggle to make accurate credit decisions, especially for customers with insufficient traditional credit histories. Conventional models, such as logistic regression, primarily depend on credit bureau data and fail to capture the full credit potential of underserved populations—thereby hindering business expansion and financial inclusion. This study investigates how NBFIs can enhance credit decision-making by applying advanced machine learning techniques—namely XGBoost and neural networks—alongside alternative data sources, including mobile phone usage patterns, utility bill payments, and social media activity. Utilizing a real-world dataset of over 300,000 individuals, the findings demonstrate that machine learning models significantly outperform traditional approaches, particularly when alternative data is incorporated. These improvements lead to more precise risk classification, enabling institutions to reduce default rates, expand lending to previously overlooked borrowers, and improve portfolio profitability. In addition, the study addresses critical ethical and privacy considerations surrounding alternative data use. The results provide actionable insights for NBFIs aiming to adopt data-driven credit strategies that balance predictive power with responsible data governance—ultimately enhancing credit operations and promoting inclusive growth.