Rule conversion in knowledge acquisition for flowshop scheduling problems

The authors examine the performance of an inductive decision tree learning system to acquire important knowledge for flowshop scheduling problems, and propose a rule conversion method from acquired rules by the system. We employ an inductive learning process for producing decision trees like the C4.5 proposed by J.R. Quinlan (1993). Several rules for job assignment are obtained from decision trees which are constructed by training cases. In the case generation method employed in the previous system, there seems to be a problem, that is, several obtained rules are not available for assigning jobs since there are no job combinations that satisfy antecedent conditions of the rules. We modify the case generation method to obtain more available rules. Computer simulations show that the modified method is effective in problems with one of the following objectives: minimizing the makespan, minimizing the total flowtime, and minimizing the total tardiness. In the previous system, only typical rules with good consequent parts had been used for job assignment. In order to utilize rules with typical bad consequent parts, we transform the antecedent parts of bad rules to try to get good rules. Computer simulations show that some bad rules can be converted to good rules