An Empirical Study About Search-Based Refactoring Using Alternative Multiple and Population-Based Search Techniques

Abstract

Automated maintenance of object-oriented software system designs via refactoring is a performance demanding combinatorial optimization problem. In this study, we made an empirical comparative study to see the performances of alternative search algorithms under. a quality model defined by an aggregated software fitness metric. We handled 20 different refactoring actions that realize searches on design landscape defined by combination of 24 object-oriented software metrics. The investigated algorithms include random, steepest descent, multiple first descent, multiple steepest descent, simulated annealing and artificial bee colony searches. The study is realized by using a tool called A-CMA developed in Java that accepts bytecode compiled Java codes as its input. The empiricial study showed that multiple steepest descent and population-based artificial bee colony algorithms are two most suitable approaches for the efficient solution of the search based refactoring problem.