Benchmark Tasks
Standardized evaluation protocols designed to measure model generalization across four key axes critical to real-world deployment of emotion sensing systems.
Generalization Tasks
TASK 01
Temporal Generalization
Evaluate how well models trained on data from earlier periods predict affective states in later time windows, capturing model robustness to temporal drift.
TASK 02
Cross-User Generalization
Evaluate model performance when applied to individuals not seen during training, assessing the degree of personalization required for effective deployment.
TASK 03
Cross-Wave Generalization
Evaluate transfer across collection waves, capturing the effect of data collection protocol changes, seasonal variation, and longitudinal behavioral shift.
TASK 04
Cross-Dataset Generalization
Evaluate model robustness when trained on one dataset and applied to another, assessing generalization across devices, demographics, and study designs.
Prediction Targets
Valence
Subjective emotional valence from positive to negative, collected via EMA self-report.
Arousal
Subjective activation level reflecting energetic arousal state at time of EMA prompt.
Stress
Perceived stress level captured through momentary self-report items in the EMA protocol.
Disturbance
Psychological disturbance or negative affect composite reflecting mental health burden.