Unveiling the Anatomy of Financial Distress Prediction in Iranian SMEs: A Hybrid Framework Coupling PLS-SEM with Explainable Machine Learning

Articles in Press

Document Type : Original Article

Authors

iliya ardeshir larijani ¹

mohammad molani aghdam ²

Miroslav Čavlin ³

¹ Department of Finance, Faculty of Management and Economics, Islamic Azad University, Science and Research Branch, Tehran, Iran.

² Innovation and Management Research Center, Am. C., Islamic Azad University, Amol, Iran.

³ Faculty of Economics and Engineering Management in Novi Sad, University Business Academy, Cvećarska 2, 21000 Novi Sad, Republic of Serbia

10.22067/ijaaf.2026.99804.1705

Abstract

The study presents an empirical approach for predicting financial distress in manufacturing small and medium-sized enterprises (SMEs) by combining Partial Least Squares Structural Equation Modeling (PLS-SEM) with supervised machine learning methods. Using a sample of 138 manufacturing SMEs from Iran, PLS-SEM was first used to check the measurement structure and to assess the relationships between the multi-dimensional risk indicators and the perceived level of financial distress. The five main predictors Debt Ratio, Market Risk, Technology Risk, Resource Risk, and Project Risk were then given to three algorithms: Artificial Neural Networks (ANN), Random Forest (RF), and Gradient Boosting (GB). The predictive procedures were refined and assessed through a nested cross-validation approach using Optuna. When the models were evaluated according to standard regression measures, ANN and RF showed very similar levels of predictive accuracy, ANN having the smallest root mean squared error (0.561) and RF performing well in terms of mean absolute error (0.446) and the coefficient of determination (0.739). In order to overcome the black-box character of the models, post-hoc interpretability was carried out using permutation feature importance and SHAP, and this showed that Debt Ratio and Market Risk made the greatest contribution to the variance in distress. The framework thus provides a transparent approach to predicting variation in Perceived Financial Distress within the observed SME setting and may support the identification of firms exhibiting stronger financial-distress signals.

Keywords

Perceived financial distress

Small and medium-sized enterprises

PLS-SEM

Machine learning

SHAP

Subjects

• Finance

---