Evolving Many-Objective Job Shop Scheduling Dispatching Rules via Genetic Programming With Adaptive Search Based on the Frequency of Features

Job Shop Scheduling (JSS) is a critical application in diverse fields, such as cloud computing and manufacturing. Genetic Programming (GP) is acknowledged for its wide use in evolving dispatching rules for JSS, offering an automated approach to heuristic generation. A dispatching rule can use many machine-related, job-related, and system-related features to create scheduling heuristics in JSS. Proper feature selection is a critical factor for the success of heuristics. Moreover, there can be many features in JSS whose importance varies from one scenario to another. It has been shown that irrelevant and redundant features can adversely affect performance. Feature selection is a promising task to select relevant features and reduce genetic programming hyper-heuristics (GPHH) search space. However, more research is needed to quantify the contribution of features in the GPHH to many-objective JSS. The proposed algorithm introduces an adaptive search strategy that is implemented through re-initialization during the evolutionary process. In addition, relevant features are selected based on their frequency of occurrence in diverse sets of best individuals. The proposed algorithm, Adaptive Feature Selection-GP-NSGA-III (AFS-GP-NSGA-III), is compared with the FS-GP-NSGA-III and standard GP-NSGA-III on a four-objective JSS problem. The experimental results indicate that feature selection using GP and adaptive search can improve the performance of the algorithm. Furthermore, the findings suggest the practical applicability of the proposed algorithm for generating improved dispatching rules for training and unseen test instances using only the selected relevant feature subset.