Learning effective new single machine dispatching rules from optimal scheduling data

• Published in: International journal of production economics 2010-11, Vol.128 (1), p.118-126
• Main Authors: Olafsson, Sigurdur, Li, Xiaonan
• Format: Article
• Language: English
• Subjects: Classification; Data mining; Decision trees; Dispatching rules; Economics; Job shops; Learning; Optimization; Production scheduling; Schedules; Scheduling; Scheduling Dispatching rules Data mining Classification Decision trees; Studies
• ISSN: 0925-5273; 1873-7579
• DOI: 10.1016/j.ijpe.2010.06.004

The expertise of the scheduler plays an important role in creating production schedules, and the schedules created in the past thus provide important information about how they should be done in the future. Motivated by this observation, we learn new scheduling rules from existing schedules using data mining techniques. However, direct data mining of scheduling data can at best mimic existing scheduling practices. We therefore propose a novel two-phase approach for learning, where we first learn which part of the data correspond to best scheduling practices and then use this data and decision tree induction to learn new and previously unknown dispatching rules. Our numerical results indicate that the newly learned rules can be a significant improvement upon the underlying scheduling rules, thus going beyond mimicking existing practice.