Ocsh: Optimized Cluster Specific Heuristics For The University Course Timetabling Problem

The University Course Timetabling Problem (UCTP) is a search problem that allocates a given number of rooms with given courses based on their scheduled time slots. UCTP belongs to the NP-complete class and can be defined as a constraint satisfaction problem (CSP). To solve the performance issue of CSP solvers, there are various local search methods. CSP solvers often use variable and value ordering heuristics to improve their search performance. Specific variable and value ordering heuristics can be even calculated by the help of a learning algorithm. Cluster-Specific Heuristics (CSH) are variable ordering heuristics which are learned based on clusters of CSPs. In this paper, to solve UCTP with user requirements, we propose a better performing CSH which is called Optimized Cluster-Specific Heuristics (OCSH). We have tested OCSH on generated time tabling problems with various user requirements and compared the runtime performances of variations of cluster-specific heuristics with OCSH. Finally, we show that OCSH is the best performing version of cluster specific heuristics to solve UCTP with user requirements.