A heuristic algorithm for the equitable partitioning problem

Abstract

This paper describes a heuristic for the equitable partitioning problem, which involves classifying individual elements, so that classes are similar. The paper presents three extensions to a heuristic algorithm, developed in earlier work, which dealt with binary-valued attributes only. The first extension illustrates how changing the coding of the data without changing the problem improves the quality of solutions obtained. The second extension allows the algorithm to deal with different scales of measurement; data sets consisting of binary valued, multi-valued nominal and interval attributes are tested and the results presented. The third extension allows the algorithm to deal with problems involving classes of different sizes. The revised heuristic is applied to the real life problem of allocating university student accommodation.