Application of data mining technology for analyzing job-shop solutions

This paper addresses the Job-Shop Scheduling Problem with Generic Time-Lags constraints (JSSP-GTL), which is a variant of the Job-Shop Scheduling Problem (JSSP). The JSSP is well known for its complexity as an NP-hard disjunctive scheduling problem. In a previous paper an efficient GRASP× ELS approach was proposed for solving JSSP-GTL. In this paper, we aim to analyze the obtained solutions using data mining technology. More specifically, considering a set of solutions that each one is obtained using a different random number generator seed, we define a distance measure between two solutions and apply a clustering technology to group them by similarity. The results show that the solutions with similar makespan are generally classified into the same cluster and they possess numerous patterns in common. Moreover, such technology could be used to define a new neighborhood, in which we consider the distance between the current solution and the incumbent solution, in order to guide the local search accepting its nearest neighbor to the incumbent solution. This paper remains on the first step toward definition of new metaheuristic which dynamically take advantages of both a clustering based optimization and pattern analysis.