Article ID: | iaor19982213 |
Country: | Netherlands |
Volume: | 88 |
Issue: | 2 |
Start Page Number: | 404 |
End Page Number: | 412 |
Publication Date: | Jan 1996 |
Journal: | European Journal of Operational Research |
Authors: | Li Der-Chiang, Lin Han-Kun, Torng Kuan-Yueh |
We explored a method of applying techniques of inductive learning from artificial intelligence to partition a full problem space into smaller exclusive problem spaces, and developed an evolving algorithm for each problem space. In this approach we first create attributes to define a problem, and use them to cluster the problem space into classes. To each class of problems, a ‘suitable’ evolved algorithm is developed to apply. By suitable here we mean that its level of complexity fits the level of difficulty of a problem of a particular type. The purpose is to increase efficiency and effectiveness. In this work we selected a developed algorithm as the parent algorithm to generate an evolved algorithm. The methods used include the technique of maximum decreasing of impurity to construct a classification tree that provides systematic class descriptions. A problem of sequencing jobs of unequal importance in a set on a single processor in order to minimize total tardiness is provided to illustrate the problem-solving procedures.