recordhulls

Recordhulls is a concept used in data analysis and computational geometry to describe the evolving set of records that lie on the convex hull of a growing collection of points. In this view, each record is represented by a point in a multidimensional attribute space, and the recordhull at a given moment consists of the points that form the boundary of the convex hull of all observed points. These points are extreme in the sense that no other combination of observed records yields a convex mixture equal to them.

In two dimensions the recordhull corresponds to the polygonal boundary, comprising the upper and lower hull; in higher dimensions it is a polytope. In a streaming or incremental setting, the hull can change as new records arrive: the new point may lie inside the existing hull (no change), lie on the hull boundary, or lie outside and cause some previously hull points to be removed from the hull.

Maintaining a dynamic recordhull uses standard convex-hull maintenance techniques. Incremental insertion with interior-point removal is common; many implementations use dynamic convex hull data structures with logarithmic amortized update times in fixed dimensions, while worst-case updates can be linear in the number of hull points.

Recordhulls are used to summarize extreme observations, support range queries and outlier detection, and aid visual analytics by highlighting the data distribution boundary. They also appear in Monte Carlo simulations and streaming analytics where the focus is on boundary behavior rather than all data points.

Because 'recordhulls' is not a universally standardized term, definitions may vary across disciplines. See also convex hull, dynamic convex hull, hull maintenance, data streams, and outliers.