Predicting SMEs’ credit risk using artificial intelligence applications: Evidence from the UK SMEs

Abstract

Financial distress is a state in which businesses struggle to pay their debts, which
frequently results in bankruptcy or company failure. Small and medium-sized
enterprises are an essential part of economies, making substantial contributions to
productivity growth, innovation and employment. Evaluating financial health is
crucial to preventing possible hardship, reducing systemic risks and ensuring long-
term stability because of its significant contribution to national and international
economies.
Bankruptcy prediction models play a critical role as early warning systems that enable
firms, lenders and policymakers to identify financial distress at an early stage and take
corrective action before failure becomes inevitable. These warning indicators may
result from internal issues, such as decreasing profitability, declining liquidity or
increasing leverage, all of which are indicative of managerial and operational
difficulties that may frequently be resolved with immediate attention.
However, External factors such as interest rate movements and fluctuations in GDP
can significantly affect firms’ operating environments by increasing borrowing costs,
suppressing demand and constraining access to finance.
The small and medium-sized business bankruptcy prediction model created by Altman
and Sabato (2007) is revisited in this thesis, which proposed two models by adding
accounting and macroeconomic data. The study further evaluates the performance of
the Altman and Sabato (2007) model by comparing results before and after the
exclusion of the retained earnings-to-total assets variable. The investigation uses data
from 2000 to 2018 and employs a variety of artificial intelligence techniques, such as
deep learning, machine learning algorithms, and ensemble methods.
The results show that macroeconomic factors greatly improve bankruptcy models'
forecast accuracy. Furthermore, the findings show that machine learning techniques
typically outperform deep learning methods in terms of accuracy. These findings
demonstrate the importance of incorporating macroeconomic variables into credit risk
assessment frameworks and have major implications for regulators, financial
institutions, business decision-makers, and academic researchers.

Date of Award	2026
Original language	English
Awarding Institution	University of Plymouth
Supervisor	Peijie Wang (Director of Studies (First Supervisor)), Alexander Haupt (Other Supervisor) & Ahmed El-Masry (Other Supervisor)