Classification research / edition 2026.10
Describe uncertain transitions
Distinguish boundary sensitivity, ambiguous states, transition estimates and calibrated probabilities.
Filtered Versus Smoothed Probabilities
Filtered and smoothed state estimates condition on different information. Neither terminology nor a sharp probability chart proves that a fitted model was available at a historical decision.
Boundaries and Abstention
A precise threshold can produce a fragile label. Measure proximity in defined units, distinguish uncertainty sources and make any decision to withhold classification explicit and auditable.
Transition Matrices When State Changes Are Scarce
Build and interpret transition counts without converting a short label history into precise forecasts, permanent states, or unsupported duration estimates.
Hysteresis, Confirmation, and the Cost of Waiting
Understand two distinct ways to stabilize classifications, including their memory, timing, missing-data rules, and limitations as evidence of a genuine state change.
Probability Calibration and What Confidence Can Mean
Define a verifiable probability target, distinguish calibration from discrimination and fitted-state confidence, and interpret small-sample scores without overstating them.