EURO 2025 Leeds
Abstract Submission

1242. Unlocking Firm Performance Insights: A Machine Learning vs. Traditional Methods Approach

Invited abstract in session WA-23: Data Analytics for Business Resilience and Sustainability - Measuring SME Performance , stream OR for Societal Development.

Wednesday, 8:30-10:00
Room: Esther Simpson 3.01

Authors (first author is the speaker)

1. Trustlord Marecha
School of Business Management, University of Johannesburg
2. Helper Zhou
School of Accounting, Economics and Finance, University of KwaZulu Natal

Abstract

Research highlights the growing need for SMEs to adopt advanced tools to understand key performance drivers. However, limited studies—especially in developing economies like South Africa—compare traditional and machine learning (ML) techniques, with the former often failing to capture complex insights. This study departs from prior empirical work that relies on linear models for policy recommendations, which have had little impact on SMEs. Using data from 1,490 South African SMEs, a quantitative approach was applied, employing Robust Regression, Generalized Least Squares (GLS), and the Enhanced Radial Basis Artificial Neural Network (K4-RANN) for comparative analysis. Findings reveal key differences: Internet Access positively influences employment in GLS but negatively in the Neural Network, suggesting automation effects. Social Media has a slight negative impact on employment in GLS but a much stronger negative effect in the Neural Network, implying reduced workforce needs. Smartphone Usage negatively impacts employment in GLS but shows a strong positive effect in the Neural Network, suggesting mobile-enabled job creation. These differences highlight how GLS captures linear trends, while the Neural Network uncovers complex, non-linear relationships. The study underscores the importance of multi-model validation in firm performance analysis, particularly for SMEs in developing economies.

Keywords

Status: accepted

Back to the list of papers