Matching Uncertain Identities Against Sparse Knowledge

This paper presents a method for fast matching of data attributes contained in a high-volume data stream against an incomplete database of known attribute values. The method is applied to vessel observational data and databases of known vessel characteristics, with emphasis on vessel identity attributes. Due to the large quantity of streaming observations, it is desirable to compute the best matching identity to a sufficient confidence level rather than include all possible identity information in the matching result. The question of which observed attributes to use in the calculation is addressed using information theory and the combination of the information conveyed by each attribute is addressed using evidence theory. An algorithm is developed which matches observations to known identities with a configurable level of desired confidence, represented as a χ value for statistical significance.