APPLICATION OF SOFT COMPUTING TECHNIQUES FOR SOFTWARE RELIABILITY PREDICTION

Abstract

Background: Software reliability prediction has become a key activity in the field of software engineering. It is the process of constructing models that can be used by software practitioners and researchers for assessing and predicting the reliability of the software product. This activity provides significant information about the software product such as “when to stop testing” or “when to release the software product” and other important information. Thus, effective reliability prediction models provide critical information to software stakeholders. Method: In this paper, we have conducted a systematic literature review of studies from the year 2005 to 2016, which use soft computing techniques for software reliability prediction. The studies are examined with specific emphasis on the various soft computing techniques used, their strengths and weaknesses, the investigated datasets, the validation methods and the evaluated performance metrics. The review also analyses the various threats reported by software reliability prediction studies and statistical tests used in literature for evaluating the effectiveness of soft computing techniques for software reliability prediction. Results: After performing strict quality analysis, we found 31 primary studies. The conclusions made based on the data taken from the primary studies indicate wide use of public datasets for developing software reliability prediction models. Moreover, we identified five most commonly used soft computing techniques for software reliability prediction namely, Neural Networks, Fuzzy Logic, Genetic Algorithm, Particle Swarm Optimization and Support Vector Machine. Conclusion: This review summarizes the most commonly used soft computing techniques for software reliability prediction, their strengths and weaknesses and predictive capabilities. The suitability of a specific soft computing technique is an open issue as it depends heavily on nature of the problem and its characteristics. Every software project has its own growth behavior and complexity pattern. Hence, more number of studies should be conducted for the generalization of the results. The review also provides future guidelines to researchers in the domain of software reliability prediction.