Enhancing Resilience of Small and Medium-Sized Enterprises in an Emerging Economy: Neural Network-Based Bankruptcy Prediction
Author:Denis Kušter and Bojana Vuković
JEL:C53, C38, G33, L26, C10.
DOI:10.24818/EA/2026/72/712
Keywords:bankruptcy prediction, neural networks, machine learning, small and medium-sized enterprises.
Abstract:
In the context of accelerated digitalisation, small and medium-sized enterprises (SMEs) increasingly operate within complex digital ecosystems that both enable growth and amplify systemic risks. Existing bankruptcy prediction models often fail to address these evolving challenges, especially in emerging markets such as Serbia. This study addresses the need for effective early-warning mechanisms by developing a data-driven bankruptcy prediction model tailored to Serbian economy. Unlike general corporate studies, this research represents a pioneering effort in the region by focusing on SMEs and integrating neural network algorithms. Utilising a balanced sample of 212 SMEs (106 solvent and 106 bankrupt), matched on key criteria such as employment, income, liabilities, and industry, the model integrated neural networks with traditional financial ratio analysis to predict bankruptcy one and two years in advance. The dataset comprised financial statements spanning from 2016 to 2022, incorporating 66 explanatory variables covering various dimensions of business performance. The findings confirmed the hypothesis, and this approach yielded superior predictive accuracy compared to established models like the
Z-score and EMS. Results demonstrated exceptional accuracy, with one-year-ahead AUC at 0.945 and the two-year-ahead model achieving an AUC of 0.835. These predictive tools serve not only as academic contributions but also as practical instruments for policymakers, financial institutions, and enterprise managers, fostering resilience and sustainable economic development in a rapidly evolving digital landscape.