A Tensor-based Approach to Nurse Rostering

Hyper-heuristics are high level improvement search methodologies exploring space of heuristics [4]. According to [5], hyper-heuristics can be categorized in many ways. A hyper-heuristic either selects from a set of available low level heuristics or generates new heuristics from components of existing low level heuristics to solve a problem, leading to a distinction between selection and generation hyper-heuristics, respectively. Also, depending on the availability of feedback from the search process, hyper-heuristics can be categorized as learning and no-learning. Learning hyper-heuristics can further be categorized into online and offline methodologies depending on the nature of the feedback. Online hyper-heuristics learn while solving a problem whereas offline hyper-heuristics process collected data gathered from training instances prior to solving the problem. Nurse rostering is a highly constrained scheduling problem which was proven to be NP-hard (Karp, 1972) in its simplified form. Solving a nurse rostering problem requires assignment of shifts to a set of nurses so that 1) the minimum staff requirements are fulfilled and 2) the nurses’ contracts are respected [3]. The problem can be represented as a constraint optimisation problem using 5-tuples: (i) set of nurses, (ii) set of days (periods) including the relevant bits from the previous and upcoming schedule, (iii) set of shift types, (iv) set of skill types and (v) constraints. In this study, a novel selection hyper-heuristic approach is employed to tackle the nurse rostering problem. The proposed framework is a single point based search algorithm which fits best in the online learning selection hyper-