STATISTICAL ANALYSIS OF OVERT AND COVERT DATA TO ASSESS LIKELIHOOD OF CBW MANUFACTURE AT A GIVEN...

MANY OBSERVABLE FEATURES CAN BE USED TO ASSESS THE LIKELIHOOD THAT A GIVEN MANUFACTURING FACILITY IS ENGAGED IN CBW PRODUCTION. MOST OF THESE FEATURES HAVE RELATIVELY SMALL DISCRIMINATION POWER, AND A MIXTURE OF ORDERED, QUALITATIVE OBSERVATIONS (ORDINAL DATA) AND QUANTITATIVE (RATIO) DATA ARE AVAILABLE. STATISTICAL ANALYSIS OF SUCH FEATURES INVOLVES COMBINING ORDINAL AND RATIO DATA SO AS TO OBTAIN A MEASURE OF WHETHER CERTAIN SITES SHOULD BE MORE CLOSELY INSPECTED FOR POSSIBLE TREATY VIOLATIONS. THE ANAYLSIS METHOD SHOULD INCORPORATE EXPERT KNOWLEDGE FROM EXPERIENCED ANALYSTS, AND THE FEATURE SET SHOULD BE EXPANDABLE TO INCLUDE NEW INSIGHTS CONCERNING ALTERNATIVE MANUFACTURING PROCESSES AND THEIR OBSERVABLE CHARACTERISTICS. A PC-BASEDPATTERN RECOGNITION SYSTEM IS TO BE CONSTRUCTED AND TESTED WITH SIMULATED OR ACTUAL OBSERVATIONS TO SOLVE THE STATISTICAL ANALYSIS PROBLEM. THE SYSTEM IS TO INCORPORATE HUMAN EXPERT KNOWLEDGE VIA FEATURE DEFINITION AND FEATURE PRIOR PROBABILITY DISTRIBUTIONS THAT CAN BE UPDATEDWITH NEW TRAINING DATA. ANTICIPATED BENEFITS/POTENTIAL COMMERCIAL APPLICATIONS - THE PATTERN RECOGNITION SYSTEM CAN BE USED FOR INDUSTRIAL MONITORING FOR ENVIRONMENTAL TREATY VERIFICATION AND NARCOTICS INVESTIGATION. IT ALSO APPLIES TO SAFETY EVALUATIONS OF LARGE SYSTEMS LIKECOMMERCIAL AIRCRAFT AND TO MARKET ANALYSIS ON THE BASIS OF OBSERVED FEATURE VALUES.