About Us | Help Videos | Contact Us | Subscriptions
 

Abstract

 

This article in AJ

  1. Vol. 81 No. 2, p. 306-312
     
    Received: Apr 22, 1988


    * Corresponding author(s):
 View
 Download
 Alerts
 Permissions
 Share

doi:10.2134/agronj1989.00021962008100020033x

AGASSISTANT: an Artificial Intelligence System for Discovering Patterns in Agricultural Knowledge and Creating Diagnostic Advisory Systems

  1. T. W. Fermanian ,
  2. R. S. Michalski,
  3. B. Katz and
  4. J. Kelly
  1. D ep. of Horticulture, Univ of Illinois at Urbana-Champaign. Urbana, IL 61801
    D ep. of Comp. Sci., George Mason Univ, Fairfax, VA 22030-4444
    D ep. of Comp. Sci., Univ of Illinois at Urbana-Champaign. Urbana, IL 61801
    B olt, Beranek, and Newman, Cambridge, MA 02238

Abstract

Abstract

AGASSISTANT is a non-specialist artificial intelligence system for automatically determining, refining, and evaluating diagnostic rules and patterns for agricultural decision problems. It has the ability to create general decision rules from examples of expert decisions. It can provide advice using these self-created rules or rules supplied by a system creator, and can also serve as a tool for developing expert or advisory systems. AGASSISTANT was first applied to build WEEDER, a system for identifying 37 grass weed or turf species commonly found in turfs in the USA. To evaluate WEEDER'S potential for exclusive identifications, a program was developed to produce all possible combinations of variable values which lead to an exclusive identification of any grass represented in the system. For most grasses, there were multiple ways ( = 4) of identifying each grass exclusively among all other grasses. Each identification required the selection of a value for an average of five variables. The maximum number of variable value decisions required for an identification was seven and the minimum was four. More than one-half (59%) of the identifications required a decision on five or less variables. WEEDER was found to be as efficient in the number of decisions required for an identification as the theoretical maximum efficiency (5 leads) for a dichotomous key covering the same species. WEEDER represents an improvement over current methods of plant identification (e.g. identification keys) due to its ability to identify specimens through multiple sets of plant variables, use uncertain knowledge, allow the user to answer questions in any order, and be easily modified to reflect local differences in plant populations. AGASSISTANT provided the total environment lor the initial representation and organization of the WEEDER knowledge base and serves as its final delivery medium.

Contribution from the Horticulture Dep., Univ. of Illinois at Urbana-Champaign. This study was supported in part by an Univ. of Illinois Res. Board grant, Int. Intelligent Systems, Inc. and was part of project no. 65-0357 of the Agric. Exp. Stn., College of Agric., Univ. of Illinois at Urbana-Champaign. Urbana, IL 61801.

  Please view the pdf by using the Full Text (PDF) link under 'View' to the left.

Copyright © .