Machine Learning Phenotyping of Autonomic Stress from Daily Temperature and Ecological Assessments

Background: Artificial intelligence applications for preventive stress monitoring remain limited by dependence on expensive continuous biosensors. We developed and validated an AI-based framework for automated phenotyping of stress-energy responses from accessible smartphone-based circadian temperature monitoring and cognitive-autonomic assessments, enabling scalable population health monitoring without wearable devices. Methods: This 15-day prospective observational study collected 239 daily observations from 16 adults (age 58.35 +/- 7.8 years; 100% adherence). Daily axillary temperature oscillation (DeltaT = night-minus-morning), a 6-item cognitive-autonomic index (MiSBIE Brief-6), morning light exposure, and screen time were analyzed using unsupervised K-means clustering. A Composite Stress Load (CSL) index integrating subjective stress (40%), thermal variance (30%), and pain (30%) was computed. Cluster validation employed silhouette analysis, Gap statistics, and Hopkins test. Results: Unsupervised machine learning identified three distinct stress-energy phenotypes (k=3; silhouette=0.75; Gap p<0.001): Cluster 1 (Low DeltaT/High Recovery; n=87; DeltaT=-0.19 +/- 0.09 degC; MiSBIE-delta=+1.84 +/- 0.62), Cluster 2 (Neutral/Intermediate; n=98; DeltaT=+0.00 +/- 0.07 degC; MiSBIE-delta=+1.12 +/- 0.51), and Cluster 3 (High DeltaT/Minimal Recovery; n=54; DeltaT=+0.21 +/- 0.10 degC; MiSBIE-delta=+0.41 +/- 0.68). Elevated DeltaT strongly correlated with CSL (r=0.52; p<0.001). AI-derived phenotypes predicted 78% of thermal stability variance (R^2=0.78; p<0.001). Morning light >15 minutes reduced DeltaT (beta=-0.24 degC; p=0.002). Conclusions: This validated AI framework achieves automated stress phenotyping at <$5 per participant versus $200-500 for wearables, supporting early identification of elevated allostatic load aligned with the Energy Resistance Principle. Longitudinal phenotype tracking enables predictive early warning and individualized exercise optimization in real-world settings, advancing health equity in preventive monitoring for resource-limited contexts. Integration into public health systems serving millions (e.g., Brazil's SUS) could enable anticipatory care delivery, improving quality of life through early intervention before clinical deterioration.