Predicting Stability of Open-Source Software Systems Using Combination of Bayesian Classifiers

Abstract

The use of free and Open-Source Software (OSS) systems is gaining momentum. Organizations are also now adopting OSS, despite some reservations, particularly about the quality issues. Stability of software is one of the main features in software quality management that needs to be understood and accurately predicted. It deals with the impact resulting from software changes and argues that stable components lead to a cost-effective software evolution. Changes are most common phenomena present in OSS in comparison to proprietary software. This makes OSS system evolution a rich context to study and predict stability. Our objective in this work is to build stability prediction models that are not only accurate but also interpretable, that is, able to explain the link between the architectural aspects of a software component and its stability behavior in the context of OSS. Therefore, we propose a new approach based on classifiers combination capable of preserving prediction interpretability. Our approach is classifier-structure dependent. Therefore, we propose a particular solution for combining Bayesian classifiers in order to derive a more accurate composite classifier that preserves interpretability. This solution is implemented using a genetic algorithm and applied in the context of an OSS large-scale system, namely the standard Java API. The empirical results show that our approach outperforms state-of-the-art approaches from both machine learning and software engineering.

A Statistical Evaluation of The Depth of Inheritance Tree Metric for Open-Source Applications Developed in Java

“…Finally, we detect the boundaries of the prediction interval for non-Gaussian data by transforming the bounds of the prediction interval for normalized data (5) using the inverse transformation (2) .…”

“…The Depth of Inheritance Tree (DIT) metric was firstly defined by Chidamber and Kemerer in [5]. The DIT metric, along with others, is used for estimating the faults [7], [19], [21], reusability [13] and other quality indicators [15] in software systems, including open-source ones [2], [8], [15], [21]. Additionally, the DIT metric is considered as a factor influencing the cost of testing [3], [20].…”

Autocorrelation Weighted Sum Entropy Based Software Quality Management for Open Source Application

“…In AWSE technique, Software reliability measures possibility of failure-free software operation using equation (8) in software program code. The Software reliability is measured in terms of percentage (%).…”

“…In [7], the evolution of Mozilla Firefox from a traditional release model to a rapid release model was analyzed to determine potential changes in field quality (users) and bug fixing (developers). A novel approach was designed in [8] for evaluating the stability of open-source software systems with aid of combination of Bayesian Classifiers. In [9], the quality evolution of an open source Java software system was examined with aid of metrics in which software quality was addressed from an internal point of view.…”

Combining text mining and data mining for bug report classification

“…The result indicates that the former combination yields a substantial improvement in accuracy over either classifier alone, and the later yields the best of all. In , Bouktif et al use the combination of Bayesian classifiers to predict the stability of open‐source software systems. The combination results in a more accurate composite classifier that preserves interpretability.…”