Credit Default Prediction for Small and Medium Enterprises Using Logistic Regression and Decision Trees

Small and Medium Enterprises (SME) form the economic backbone of global enterprise
ecosystems, yet evaluating their creditworthiness presents significant challenges for
commercial lending institutions due to opaque financial disclosures, non-linear risk profiles,
and macroeconomic vulnerabilities. This paper develops a rigorous empirical benchmark
comparing parametric Logistic Regression and non-parametric Decision Tree models for
SME credit default prediction. Grounded in non-linear risk estimation architectures and
SME Enterprise Risk Management (ERM) frameworks, our methodology incorporates high
dimensional accounting ratios, liquidity deltas, and macro-financial indicators across an
empirical dataset of 65,000 SME loan portfolios observed between 2018 and 2024. The
empirical results reveal that while Logistic Regression delivers robust interpretability and
log-odds parameter stability (AUC-ROC = 0.812), Decision Trees capture critical non-linear
interaction effects and threshold cutoffs (AUC-ROC = 0.865), reducing Expected Credit
Loss (ECL) by 24.8%. Integrating both models into an optimal hybrid voting ensemble
yields an AUC-ROC of 0.894 and a True Positive Rate (TPR) of 88.2% at a 0.15
Significance Level. We present an exhaustive quantitative synthesis incorporating 10 data
tables, 10 mathematical behavior graphs, and 20 distinct chart architectures detailing risk
capital allocation, feature importance, and portfolio loss mitigation dynamics.