Book Description

"As computers are increasingly relied upon to perform tasks of increasing complexity affecting many aspects of society, it is imperative that the underlying computational methods performing the tasks have high performance in terms of effectiveness and scalability. A common solution employed to perform such complex tasks are computational intelligence (CI) techniques. CI techniques use approaches influenced by nature to solve problems in which traditional modeling approaches fail due to impracticality, intractability, or mathematical ill-posedness. While CI techniques can perform considerably better than traditional modeling approaches when solving complex problems, the scalability performance of a given CI technique alone is not always optimal. Hybridization is a popular process by which a better performing CI technique is created from the combination of multiple existing techniques in a logical manner. In the first paper in this thesis, a novel hybridization of two CI techniques, accuracy-based learning classifier systems (XCS) and cluster analysis, is presented that improves upon the efficiency and, in some cases, the effectiveness of XCS. A number of tasks in software engineering are performed manually, such as defining expected output in model transformation testing. Especially since the number and size of projects that rely on tasks that must be performed manually, it is critical that automated approaches are employed to reduce or eliminate manual effort from these tasks in order to scale efficiently. The second paper in this thesis details a novel application of a CI technique, multi-objective simulated annealing, to the task of test case model generation to reduce the resulting effort required to manually update expected transformation output"--Abstract, page iv.