Segmentun

Segmentun is a unit of analysis representing a minimal contiguous segment within a dataset that satisfies a defined homogeneity criterion. It is used in sequential data analysis to allow efficient summarization and comparison across segments.

In practice, a segmentun is defined as the smallest contiguous region such that any further subdivision would violate a chosen property, e.g., a maximum within-segment variance or a bounded change-point. Because criteria vary, there is no universal canonical definition; different domains may specify different thresholds.

Computation often relies on dynamic programming, thresholding, or hierarchical methods. Common approaches identify change points and then merge adjacent segments if the merged region still satisfies the criterion, or split if not.

Applications include time series analysis, speech recognition, genomic annotation, and image processing where sequential or spatial data is partitioned into meaningful units. An example is a time series with segments of near-constant mean, where segmentuns are the minimal blocks with stable mean within a tolerance.

The concept has varying implementations and is sometimes referred to by different names such as change-point blocks or homogeneous segments. Evaluation depends on domain-specific validation, and researchers caution against over-interpretation of segmentun boundaries in noisy data.