Skip to main content
U.S. flag

An official website of the United States government

Development and Evaluation of an Expert System for Diagnosing Pest Damage of Red Pine

Informally Refereed

Abstract

An expert system for diagnosing pest damage of red pine stands in Wisconsin, PREDICT, runs on IBM or compatible microcomputers and is designed to be useful for field foresters with no advanced training in forest pathology or entomology. PREDICT recognizes 28 damaging agents including species of mammals, insects, and pathogens, as well as two types of abiotic damage. Two separate development tools (EXSYS and INSIGHT2+) were used. Each employs a rule-based method for representing knowledge, which was obtained from the literature and from human experts in the fields of forest pathology and entomology. The pest-inference rule blocks, for each damaging agent, are based on the abduction model of diagnosis and include rules for eliminating a pest from further consideration, diagnosing a pest as certain, and compiling evidence in favor of a pest. Both development tools employ a backward-chaining control strategy; however, it was necessary to modify this approach by designing a special block of rules to approximate the mixed strategy used by the human experts. A logic and completeness rule block was also constructed to deduce facts omitted by the user and to minimize the need for questioning. Input to PREDICT is obtained from pest damage reports containing specific information about stand/site conditions, tree symptoms, and signs. Diagnoses from PREDICT take the form of a list of one or more possible agents with corresponding confidence values. Actual and hypothetical test cases were used to refine the knowledge base, then a separate set of 20 actual cases was used as a basis for testing and evaluating the completed system. It was necessary to develop special procedures for refining and evaluating the system to accommodate the often vague and uncertain nature of pest damage information. Two versions of PREDICT (developed with the EXSYS and INSIGHT2+ tools, respectively) were evaluated and compared with three recognized experts and two field foresters. No significant differences were found between the performances of PREDICT and the experts; however, PREDICT performed significantly better than the two foresters, even though they both have training and experience in forest pest diagnosis. It was concluded that PREDICT is able to improve the diagnoses of field foresters to a level comparable with recognized experts.

Citation

Schmoldt, Daniel L; Martin, George L. 1989. Development and Evaluation of an Expert System for Diagnosing Pest Damage of Red Pine. Forest Science. June, 1989: 364-387.
https://www.fs.usda.gov/research/treesearch/116