Interpretable Multi-Model Framework for Early Warning of SME Loan Delinquency
Preprint
- 27 November 2025
- preprint
- Published by MDPI AG in Preprints.org
Abstract
The rapid expansion of small and medium enterprise (SME) lending has intensified the need for accurate and interpretable credit risk forecasting. Financial institutions must an-ticipate potential business loan delinquency to maintain portfolio stability and meet reg-ulatory standards. This study proposes a multi-model analytical framework that inte-grates statistical, probabilistic, and machine learning methods—Linear Regression, Naïve Bayes, and NeuralProphet—to improve both interpretability and predictive accuracy in delinquency forecasting. Linear Regression is employed to identify key macroeconomic and portfolio-level variables that influence default behavior, providing transparent and explainable insights with statistically significant coefficients. The Naïve Bayes model in-troduces a probabilistic layer to assess the likelihood of delinquency escalation under var-ying business conditions. Finally, the NeuralProphet model performs nonlinear time-series forecasting using the selected regressors and achieves an R² of 0.89 with a mean absolute percentage error of 8.5 percent, while residuals remain approximately normally distributed according to the Shapiro–Wilk test. The results show that the pro-posed multi-stage framework effectively combines interpretability and predictive perfor-mance, supporting explainable, data-driven decision-making in SME credit risk manage-ment. The approach contributes to the development of intelligent forecasting and control systems for modern financial institutions.Keywords
All Related Versions
- Version 1, 2025-11-27, Preprints
- Published version: Risks, 14 (2), 25.
This publication has 42 references indexed in Scilit:
- Probabilistic forecasting with VAR-VAE: Advancing time series forecasting under uncertaintyInformation Sciences, 2025
- Forecasting macroeconomic data with Bayesian VARs: Sparse or dense? It depends!International Journal of Forecasting, 2025
- Bayesian hierarchical probabilistic forecasting of intraday electricity pricesApplied Energy, 2024
- Deep learning-based time series forecastingArtificial Intelligence Review, 2024
- Hybrid time series models with exogenous variable for improved yield forecasting of major Rabi crops in IndiaScientific Reports, 2023
- Probabilistic time series forecasts with autoregressive transformation modelsStatistics and Computing, 2023
- Forecasting commodity prices: empirical evidence using deep learning toolsAnnals of Operations Research, 2023
- Neural basis expansion analysis with exogenous variables: Forecasting electricity prices with NBEATSxInternational Journal of Forecasting, 2022
- M5 accuracy competition: Results, findings, and conclusionsInternational Journal of Forecasting, 2022
- A state space framework for automatic forecasting using exponential smoothing methodsInternational Journal of Forecasting, 2002