Application of artificial immune systems in software fault prediction problem

Because of rising complexity in software systems and increasing user expectations, it is necessary to manage the software quality as an engineering discipline called ?Software Quality Engineering?. Software fault prediction is one of the quality assurance activities which locates in Software Quality Engineering discipline. In this thesis study, the software fault prediction problem has been tried to be solved with Artificial Immune Systems (AIS) based algorithms and new models have been proposed by examining existing algorithms. The performance of different AIS based classifiers has been examined on five NASA datasets and compared with the best machine learning algorithms reported in literature. With the help of this study, AIS based classifiers? performance have been investigated for the first time in literature. Thanks to the high performance of AIRS algorithm, an AIRS based model using ?correlation-based feature selection? technique has been proposed. This model specifically provided high performance for very large datasets and results were better than many learning algorithms. If a similar model can be built using class-level metrics, fault-prone modules can be identified and refactored in design phase. By using this idea, experimental analysis have been conducted using class-level Chidamber-Kemerer (CK) metrics on fault prediction models. It has been empirically proved that depth of inheritance tree (DIT) is the least significant metric and coupling between object classes (CBO) is the most significant one for the fault prediction. It has been reported that AIRS based prediction model using CK metrics and lines of code provides the highest performance.Semi-supervised learning is one of the most active research topics in machine learning. An AIS based semi-supervised learning algorithm has been proposed and applied in this problem. It has been determined that AIRS algorithm?s performance increases with the proposed algorithm when there is limited fault data. However, this approach decreases the performance of some algorithms for software fault prediction. In addition, a framework to use the software fault prediction in ?Software Product Lines? approach, which is popular since the beginning of the 2000s for software reusability, has been proposed. In addition to AIS based fault prediction models, ?prediction-centric software life cycle? and ?prediction-centric software processes? have been proposed as a new development approach and a new process approach. Furthermore, that study showed how to use the fault prediction approaches to prioritize system test cases and emphasized the benefits of fault prediction approaches.