Scatter search for solving the course timetabling problem

Abstract

Scatter Search (SS) is an evolutionary population-based metaheuristic that has been successfully applied to hard combinatorial optimization problems. In contrast to the genetic algorithm, it reduces the population of solutions size into a promising set of solutions in terms of quality and diversity to maintain a balance between diversification and intensification of the search. Also it avoids using random sampling mechanisms such as crossover and mutation in generating new solutions. Instead, it performs a crossover in the form of structured solution combinations based on two good quality and diverse solutions. In this study, we propose a SS approach for solving the course timetabling problem. The approach focuses on two main methods employed within it; the reference set update and solution combination methods. Both methods provide a deterministic search process by maintaining diversity of the population. This is achieved by manipulating a dynamic population size and performing a probabilistic selection procedure in order to generate a promising reference set (elite solutions). It is also interesting to incorporate an Iterated Local Search routine into the SS method to increase the exploitation of generated good quality solutions effectively to escape from local optima and to decrease the computational time. Experimental results showed that our SS approach produces good quality solutions, and outperforms some results reported in the literature (regarding Socha's instances) including population-based algorithms.