Kranti Nebhwani Undergraduate Dissertation 2015/16

Cellular Genetic Algorithms for Test Generation

Supervised by G.Fraser

Abstract

Currently software testing is a crucial and labor-intensive component of software production. Software testing requires test data generation with good code coverage. Automated search techniques used for software test generation will be an improvement to efficiency and cost to software testing, and currently Genetic Algorithms are one the best options as a search heuristic for automated test generation. The project will investigate in Cellular Genetic Algorithms, another subclass of Evolutionary Algorithms just like Genetic Algorithms, and their potential to replace Genetic Algorithms for test generation with even better code coverage. Research and implementation were done in regards to how these techniques may be applied to test generation, and specifically how using Cellular GAs may improve upon using standard GAs for test generation. The tests were done with the software tool Evosuite which runs a GA on Java classes to generate test-suites and outputs the GA’s performance on code coverage, and a Cellular GA will be implemented into EvoSuite in this project and be compared to the standard GA in its performance for generating test data. Results achieved showed general favor for cellular GAs due to the advantage of creating more diverse populations than standard GAs and also prevent premature convergence of solutions that more often converged at a local optimum instead of branching out searching for optimal solution for the standard GA.