A common approach to precision medicine is stratifying individuals into homogeneous subgroups based on defined criteria. In previous studies, infrared molecular fingerprinting (IMF), a blood-based profiling method that captures broad molecular information, has been proposed for personalized health monitoring due to its low intra-individual variability relative to the population-level variability. In a personalized setting, deviations from the healthy state may thus be more sensitively captured. To enable this in practice, subgroups with reduced interpersonal variability must be identified. This study explores the existence, prediction, and explainability of such subgroups within IMFs. Using a cohort of 4032 healthy individuals with up to 5 visits each, we show that subgroups of reduced interpersonal variability exist. The first three principal components (PCs) of the high-dimensional IMFs are sufficient to define subgroups where within-subgroup variability approaches the levels of intra-individual variability. Machine learning models are trained to predict these PCs from routine clinical chemistry, IMF measurement parameters, and participants’ characteristics (demographics, lifestyle, and health-related variables). The PCs are successfully predicted with inaccuracies close to or below intra-individual variability. Using Shapley Additive Explanations, we identify key factors behind subgroup formation, ensuring interpretability.