An investigation of the relationship between problem characteristics and algorithm performance: a case study of the GAP

Abstract

Abstract We compare synthetic Generalized Assignment Problems (GAP) generated under two correlation-induction strategics: Implicit Correlation Induction (ICI) and Explicit Correlation Induction (ECI). We present computational results for two commercially-available solvers and four heuristics on 590 test problems. We conclude that the solvers’ performances degrade as the population correlation between the objective function and capacity constraint coefficients decreases. However, the heuristics’ performances improved as the absolute value of the same population correlation increases. We find that problems generated under ECI arc more challenging than problems generated under ICI and may lead to a better understanding of the capabilities and limitations of the solution methods being evaluated. We recommend that one consider the purpose(s) of an experiment and types of solution procedures to be evaluated when determining what types of test problems to generate.