Abstract
Nowadays, software systems evolve in vast and complex applications. In such a complex system, a minor change in one part may have unexpected degradation of the software system design, leading to an unending chain of bugs and defects. Therefore, to keep track of implications that could appear after a change has been applied, the assessment of the software system is of utmost importance. As a result, in this direction, software metrics are suitable for quantifying various aspects of system complexity and predicting as early as possible those parts of the system that could be error-prone. Thus, in this paper, we propose a comparative study of two complexity metrics, Weighted Method Count and Hybrid Cyclomatic Complexity, regarding the prediction of software defects. Specifically, the objective is to investigate whether using a hybrid metric that measures the complexity of a class improves the performance of the fault prediction model. We conduct a series of several experiments on five open source projects datasets. The preliminary results of our research indicate that the proposed metric performs better than the standard complexity metric of a class, Weighted Method Count. Moreover, the Hybrid Cyclomatic Complexity metric can be seen as a base for building a more complex and robust complexity metric.
Citare
@Inproceedings{Cernau2022AHC,
author = {L. Cernau and L. Dioşan and C. Serban},
booktitle = {International Conference on Software and Data Technologies},
title = {A Hybrid Complexity Metric in Automatic Software Defects Prediction},
year = {2022}
}