A Theoretical Formulism for Evidential Reasoning and Logic Based Bias Reduction in Geo-Intelligence Processing

Abstract

- Geo-intelligence processing is strongly based on the need to bring together analytical viewpoints from multiple members comprising a geo-intelligence team so that unified answers to problems can be provided to leadership responsible for decision making. A three-tier evidential reasoning formulism is proposed and explained embodying a guide for the statistical/cognitive processing of geo-intelligence sensor information to facilitate this aim. The first tier comprises computational modeling used in conjunction with informal logic-based bias reduction by a multiple analyst team to interpret geo-intelligence information and create geo-intelligence reports. In the second tier, Bayesian belief networks over distinct provinces under geo-intelligence analytical investigation are created by each analyst through the amalgamation of statistical information provided by geo-intelligence reports. Bayesian belief network (BBN) results coupled with ancillary intelligence and analyst beliefs provide a set of propositions and probability masses summarizing the state of each province analyzed by each team member. The BBN state levels denote the three conditions of lack of nefarious substance presence, probable nefarious substance presence, and definite nefarious substance presence and are taken to be related, via a one-to-one mapping, directly to a new set of decision-based propositions – lack of adversary attack, probable adversary attack, and definite adversary attack. In the third tier, team member probability masses associated with these propositions, along with conjunctive and disjunctive combinations, are gradually amalgamated using Dezert-Smarandache (DS) evidential theory. A numerical example demonstrates the mechanics of the third-tier information fusion process which takes into account logical paradoxes and results in a single virtual analyst probability mass distribution associated with the geo-intelligence information amalgamation problem.