Unmasking indoor exposure profiles in vulnerable households: an unsupervised clustering of integrated environmental and demographic data

Indoor air quality (IAQ) represents a complex interplay of exposures that single-pollutant studies often fail to capture. Identifying holistic exposure profiles is critical for understanding health risks in populations spending the majority of their time indoors. We integrated high-resolution environmental sensor data with demographic and behavioral metadata from residential testbeds. To handle the high-dimensional, heterogeneous data, we first engineered temporal and health-relevant categorical variables. We then applied Factor Analysis of Mixed Data (FAMD) followed by Uniform Manifold Approximation and Projection (UMAP) for dimensionality reduction. On this reduced feature space, we employed Torque clustering, a parameter-free algorithm, to identify stable exposure patterns. The resulting profiles were characterized using heatmaps and Z-score-based radar charts, and statistically validated through odds ratio (OR) analysis. The analysis identified six stable and distinct exposure profiles that remained consistent across different time periods. Key profiles included an acoustic-dominated group (Cluster 0), hot-and-humid physical environment groups (Clusters 1 & 2), and chemical-dominant groups (Clusters 4 & 5), all compared against a low-exposure reference group (Cluster 3). Specifically, Cluster 2, characterized by an older adult demographic, represented a ’multi-hazard’ group with the highest relative risk for adverse TVOCs levels (OR ≈ 50). These profiles were significantly associated with demographic and behavioral factors, including age, housing type, and gas stove use. This study demonstrates that unsupervised clustering can systematically stratify heterogeneous indoor exposures into robust, interpretable profiles. By moving beyond single-pollutant assessment, this data-driven approach provides a powerful framework for precision environmental health strategies by enabling cluster-specific risk stratification and targeted interventions.