Learning dispatching rules for single machine scheduling with dynamic arrivals based on decision trees and feature construction

Abstract

In this paper, we address the dynamic single-machine scheduling problem for minimisation of total weighted tardiness by learning of dispatching rules (DRs) from schedules. We propose a decision-tree-based approach called Generation of Rules Automatically with Feature construction and Tree-based learning (GRAFT) in order to extract dispatching rules from existing or good schedules. GRAFT consists of two phases: learning a DR from schedules, and improving the DR with feature-construction-based genetic programming. With respect to the process of learning DRs from schedules, we present an approach for transforming schedules into training data containing underlying scheduling decisions and generating a decision-tree-based DR. Thereafter, the second phase improves the learned DR by feature-construction-based genetic programming so as to minimise the average total weighted tardiness. We conducted experiments to verify the performance of the proposed approach, and the results showed that it outperforms the existing dispatching rules. Moreover, the proposed algorithm is effective in terms of extracting scheduling insights in such understandable formats as IF–THEN rules from existing schedules and improving DRs by grafting a new branch with a discovered attribute into a decision tree.

Keywords:

• Scheduling
• single-machine scheduling
• decision tree
• genetic programming
• machine learning
• feature construction
• dispatching rules

Disclosure statement

No potential conflict of interest was reported by the author(s).

Correction Statement

This article has been corrected with minor changes. These changes do not impact the academic content of the article.