E1.S5

DATASET DESCRIPTION

Disease Area

Behavioral Health — Sleep Disorders & Major Depressive Disorder (MDD)

Specific Research Focus

Sleep disruption classification and MDD risk prediction in young adults aged 18–30, with longitudinal treatment response modeling

Dataset Theme / Edition

Behavioral Health & Sleep Dataset — Edition 1.0 (SomniMetrics-BHDS v1.0)

Short Dataset Summary

A high-fidelity synthetic dataset of 1,000 young adult patients combining cross-sectional sleep architecture metrics, psychometric scale scores (PHQ-9, GAD-7, PSS-10, PSQI, ESS), wearable biomarkers (HRV, sleep efficiency, REM%), and behavioral lifestyle features — paired with 12,073 longitudinal follow-up session records across six clinical trajectory archetypes

Intended Research Applications

Sleep disorder classification, MDD early detection from sleep biomarkers, CBT-I treatment response prediction, wearable-based behavioral anomaly detection, stress-sleep-mood cascade temporal modeling, XAI/SHAP attribution research, and federated learning benchmarking in digital mental health

Total Number of Patients

1,000 synthetic patients (cross-sectional) with 12,073 total longitudinal session rows, averaging 12.1 sessions per patient

Number of Features / Parameters

35 cross-sectional features + 26 longitudinal session-level features = 61 total parameters across both dataset parts

Data Modalities Included

Tabular clinical assessments, validated psychometric scales, wearable-derived biosignals (HRV, sleep efficiency, REM%), self-reported behavioral and lifestyle variables, and structured longitudinal time-series session records

Recommended AI Tasks

Multi-class classification (4-class sleep disruption severity), regression (PHQ-9 score, relapse risk 0–1), sequence modeling (LSTM, Transformer), anomaly detection, survival analysis (time to remission), and conditional generative modeling (TimeGAN-CR)

Dataset Size

Part A CSV: ~176 KB · Part B CSV: ~1.5 MB · NumPy tensor (1000 × 17 × 21): ~1.4 MB · Total across all files: ~5 MB

Release Date

May 2026

Why This Dataset Exists

Privacy constraints make large, well-labeled behavioral health datasets with wearable biomarkers and longitudinal follow-up nearly impossible to share openly. SomniMetrics-BHDS was created to give the research community a clinically rigorous, privacy-safe training and benchmarking resource that no IRB barrier can block

Real-World Problem Addressed

Sleep disruption affects 30–40% of young adults and is the single strongest modifiable predictor of MDD onset and relapse — yet AI tools for early detection remain underdeveloped due to data scarcity. This dataset directly targets that gap by enabling model development for the population most at risk and least served by existing clinical AI infrastructure