Derivation of a Reliability Metric for Fused Data Decision Making

Users require information fusion to reduce dimensionality for real world, complex decision-making. Typically, researchers design fusion systems based on limited data that does not capture all operating conditions seen in the real world (i.e. weather for video sensors). Fusion systems are of limited use when presented with poor data, inappropriate models, and unrealistic assumptions. Decision makers are burdened with the task of determining the quality of fused output based on trial and error. If the fusion system works in most scenarios, aids the user in purposeful decisions, and leady to successful actions; the user has high system confidence. However, if the fusion system causes erroneous results that lead to poor decisions, the user disposes of the entire system - which could have been the result of bad data, incomplete models, and restricted assumptions. Thus, when a fusion system is fielded, there has to be metrics associated with model fidelity, data uncertainty, and constraints over applicability. By combing representations of data quality; this paper derives a reliability metric to aid users to trust fusion outputs, perform a utility assessment, and refine sensor collections