A Software Topology Based Concern Point Metric

It is essential for software development and maintenance to predict software defects in an object-oriented (OO) software system.(The results can serve as the basis for workload scheduling and arrangements, in order to improve the efficiency of software development and maintenance. Software defects prediction uses a variety of metrics to analyze data, build models and predict software defects in the project. In order to continuously improve the performance of software defects prediction models, there are many specific metrics have been proposed in these years, such as cohesion and coupling.In order to further improve the performance of the traditional software defects prediction models, we proposed a software topology based metric for predicting fault-prone classes, named "concern point". For a given object-oriented system, first we use the software topology of all classes to build a dependency graph. Then, we calculate the values of "concern point" metric for all classes in the basis of the dependency graph we built. Finally, we collect data of three Eclipse versions (2.0,2.1and3.0) and compare four specific models for predicting software defects in the experiment. Our results indicate that, on one hand, the proposed metrics capture different information from traditional OO metrics. On the other hand, the proposed "concern point" metric can significantly improve the performance of the traditional software defects prediction models.The main contributions of this thesis are summarized as follows: (1) We present a dependency graph to capture the topology between classes and give the method to construct the dependency graph;(2) We propose a software topology based metric for predicting fault-prone classes, in order to improve the performance of the traditional software defects prediction models.(3) We empirically investigate the effectiveness of the proposed metric.