Testing of heuristic methods: A case study of greedy algorithm

Abstract

Algorithms which seek global optima are computationally expensive. Alternatively, heuristic methods have been proposed to find approximate solutions. Because heuristic algorithms do not always deliver exact solutions it is difficult to verify the computed solutions. Such a problem is known as the oracle problem. In this paper, we propose to apply Metamorphic Testing (MT) in such situations because MT is designed to alleviate the oracle problem and can be automated. We demonstrate the failure detection capability of MT on testing a heuristic method, called the Greedy Algorithm (GA), applied to solve the set covering problem (SCP). The experimental results show that MT is an effective method to test GA. © 2011 IFIP International Federation for Information Processing.