Behavior learning engines (BLEs) are platforms intended to enable the discovery, visualization and analysis of recurring, complex, multiperiod patterns in large operational performance data sets. If such engines are to realize their intent, they must support four layers of cumulative functionality:

Variable designation—this supports the selection of the system properties that are to be tracked by the platform and how those properties are to be measured

Variable value normalization&mdahs;this is the automated ability to determine (usually by an algorithm that regresses measurements) what constitutes the normal or usual values assumed by the system property measuring variables

Observational dependency modeling—this is a set of tools for linking the individual property measuring variables to one another, where the links represent some kind of dependency among the values taken by the linked variables; commercially available BLEs differ significantly with regard to the degree to which the dependencies must be pre-established by the vendor or user and the degree to which the dependencies are themselves discovered by BLE algorithms working on the performance data sets being considered

Assessment—the means by which, once the normalized values and dependency map are determined, the resulting construct can be used to answer questions about the values assumed by some variables, based on the values assumed by others