An unsupervised machine learning analysis of stakeholder opinions on foot-and-mouth disease control in South Africa

Abstract

Foot-and-mouth disease (FMD) remains one of the most contagious and economically disruptive livestock diseases in South Africa, with recurrent outbreaks affecting animal health, wildlife conservation, and national trade. Effective control planning requires not only epidemiological evidence but also an understanding of how proposed interventions are perceived by different stakeholder groups. This study analyzed survey-based feasibility assessments of key FMD control strategies, including vaccination, culling, fencing, and movement control, collected from expert stakeholders with professional experience in animal health and members of the general public in South Africa. Using an unsupervised machine learning framework, principal component analysis was applied to reduce the high-dimensional survey data into interpretable latent components, followed by K-means clustering to identify groups with similar feasibility profiles. Unlike conventional decision-analytic approaches that emphasize average scores, this approach revealed latent opinion structures within and across stakeholder groups. The analysis identified three distinct expert clusters characterized by differing strategic orientations, while public respondents exhibited greater heterogeneity and weaker cluster separation. These findings highlight both consensus areas and systematic divergences between stakeholder groups, offering empirically grounded insights to support inclusive, context-sensitive, and evidence-informed FMD control policy design in South Africa.