An assessment of a days off decomposition approach to personnel shift scheduling

This paper studies a two-phase decomposition approach to solving the personnel scheduling problem. The first phase creates a days-off-schedule, indicating working days and days off for each employee. The second phase assigns shifts to the working days in the days-off-schedule. This decomposition is motivated by the fact that personnel scheduling constraints are often divided into two categories: one specifies constraints on working days and days off, while the other specifies constraints on shift assignments. To assess the consequences of the decomposition approach, we apply it to public benchmark instances, and compare this to solving the personnel scheduling problem directly. In all steps we use mathematical programming. We also study the extension that includes night shifts in the first phase of the decomposition. We present a detailed results analysis, and analyze the effect of various instance parameters on the decompositions’ results. In general, we observe that the decompositions significantly reduce the computation time, but the quality, though often good, depends strongly on the instance at hand. Our analysis identifies which aspects in the instance can jeopardize the quality.