Abstract
Present wellness monitoring systems primarily emphasize single-modality metrics, neglecting the intricate, interdependent interactions among cardiac, neural, and behavioral regulatory systems. This paper introduces a machine learning framework that implements heart–brain–body coherence as a dynamic, longitudinally evaluated wellness metric. In contrast to previous simulation-based studies that utilized independently generated, cross-sectional data categorized by deterministic IF-THEN rules, the current study adopts a longitudinal simulation design featuring 80 virtual subjects monitored across 30 consecutive daily profiles (N=2,400 records). This design incorporates physiologically realistic inter-variable correlations and probabilistic coherence labelling derived from a continuous risk-scoring function. The subject-level train/test split (56 training subjects, 1,680 records; 24 test subjects, 720 records) stops data from leaking over time. The precision for logistic regression was 79.2% (ROC-AUC = 0.887; five-fold CV AUC=0.904±0.010), while random forest's precision was 78.1% (ROC-AUC = 0.871; five-fold CV AUC=0.894±0.010). The values were significantly varied across all inputs, with an HRV RMSSD of 22.6%, self-reported fatigue with 21.2%, hours of sleep with 19.9%, and stress score at 18.7%. This shows real multi-signal coherence instead of artefacts that are caused by variables that create labels. These results establish a reliable simulation benchmark for subsequent validation with authentic wearable datasets, including MIMIC-IV and PhysioNet.