Table 1_Unsupervised discovery of public health subgroups associated with cognitive health vulnerability in adults: an analysis of BRFSS 2024.docx

However, latent population heterogeneity embedded in routine public health survey variables remains insufficiently explored.</p>Objective<p>This study aims to develop an unsupervised semantic stratification framework based on BRFSS 2024 data to identify adult population subgroups associated with cognitive health-related vulnerability from a public health perspective.</p>Methods<p>Routine public health variables were selected, cleaned, imputed, and standardized, while cognitive reference variables were excluded from model training.

An autoencoder was used to learn low-dimensional latent representations, followed by MiniBatchKMeans clustering to identify semantic subgroups. Each subgroup was further analyzed through public health profiling, held-out cognitive reference validation, non-redundant feature signature analysis, and intervention priority assessment.</p>Results<p>Four interpretable semantic subgroups were identified: the low-burden subgroup, general health vulnerability subgroup, health behavior risk subgroup, and preventive service gap subgroup.