Identifying health risk profiles among older adults in Kerala using machine learning

Multidimensional clustering, predictive determinants, and public health implications

Bibliographic Data

ID22091601
AuthorsBlessy Sarah Mathew (0000-0003-4750-5463, Lovely Professional University), Olusiji Adebola Lasekan (0000-0002-4244-4852, Universidad Católica de Temuco), Margot Teresa Godoy Pena (0009-0009-7480-7185, Universidad de La Frontera), Rekha K (Lovely Professional University), Swaty Sharma (Lovely Professional University), S R Krupesha Sharma (0000-0001-6596-3834, Lovely Professional University), Salini B Nair (Elijah House)
Year2026
Volume14
Pages1783083-1783083
Publication date2026-04-17
Open AccessYes
TypeARTICLE
VenueFrontiers in Public Health (JOURNAL)
Journal identifiersISSN: 2296-2565 • E-ISSN: 2296-2565
PublisherFrontiers Media SA (PUBLISHER • CH)
DOI10.3389/fpubh.2026.1783083
PMID42077956
OpenAlexW7154708482
LanguageEN
References cited58

This study aimed to identify multidimensional health risk profiles in older adults using an integrated machine learning framework that combines unsupervised clustering, dimensionality reduction, and supervised classification. A cross-sectional survey of 800 community-dwelling adults aged 60 years and above in Kerala was analyzed using K-Means clustering based on indicators of digital health literacy, functional difficulty, and self-management capacity, followed by Principal Component Analysis (PCA) to visualize multidimensional cluster separation. A Random Forest classifier was applied to assess the internal consistency and demographic alignment of the identified clusters, and non-parametric statistical tests were applied to assess between-cluster differences in health and digital indicators. The analysis identified two distinct clusters representing low-risk and high-risk profiles in older adults, with good internal validity (average Silhouette score = 0.30). The identified clusters are interpreted as descriptive profiles derived from exploratory unsupervised analysis rather than causal groupings. The high-risk group was characterized by substantially lower digital health literacy, reduced trust in online health information, greater functional difficulties, and poorer self-management capacity, rather than chronic disease presence alone. Statistical testing confirmed significant differences between clusters across digital health abilities, functional indicators, and selected health characteristics. Education, income, and age were most strongly associated with cluster membership, while living arrangement and employment status showed moderate association. Overall, the findings demonstrate that health risk in older adults is inherently multidimensional and associated with the interaction of digital capability, functional health, and socioeconomic position rather than by demographic factors alone. This machine learning–driven approach provides empirically grounded insights for targeted, equity-oriented interventions and supports more precise older care planning in Kerala’s rapidly ageing context

Cluster analysis · Digital health · Linear discriminant analysis · Principal component analysis · Psychological intervention · Public health · Random forest · Silhouette · Chronic Disease Management Strategies · Frailty in Older Adults · Technology Use by Older Adults