Interpretation & discovery
Longitudinal analysis
Understand how Ounias derives transitions, temporal summaries, and trajectory information from repeated observations.
Overview
Longitudinal analysis extends a labeling result across repeated observations belonging to the same entity.
It transforms sequences of analytical states into transition, timing, dwell, movement, and trajectory-level information while preserving their relationship to the parent run.
Dataset and parent-run requirements
- An entity identifier
- Repeated observations per entity
- An observation index or observation datetime
- A compatible parent labeling result
Analytical outputs
- Label sequences
- Start and terminal state profiles
- Transition profiles
- Jump counts
- Temporal profiles
- Dwell profiles where datetime information is available
- Movement and directionality information
- Trajectory-level representations
Supported analytical contexts
- Classification — Follow discrete supervised states across repeated observations.
- Regression — Represent continuous outcomes as finite states when trajectory analysis requires categorical transitions.
- Clustering — Follow unsupervised labels through repeated observations in feature or embedded analytical space.
Interpretation and limitations
- Longitudinal results inherit assumptions and uncertainty from the parent labeling model.
- Observed state changes should not automatically be interpreted as causal or clinically meaningful transitions.
- Temporal interpretation depends on the quality and meaning of observation ordering or datetime information.