SOFTWARE PROCESS IMPROVEMENT USING DEFECT PREDICTION ANALYTICS

Abstract

Defects in software cause additional time and financial costs for businesses, and can result in significant delay in completion of software projects, customer not satisfied with software products, and a multiple of other problems. Most industrial software systems find the majority of their defects later in the software development life cycle than they should be. At this point in time the cost of fixing those defects is significantly increased compared to if they had been found earlier in the development process, and the quality assurance processes used are less effective. Previous studies have indicated that the quality assurance processes that focus on testing are largely reactive and do not provide a valid means of ensuring quality assurance in complex software systems.

In order to mitigate these problems, this research uses defect prediction analytics to improve software development processes. Specifically, the proposed research utilizes historical software metric data along with machine learning algorithms to identify defect-prone modules in the software development process at a relatively early stage in the software development lifecycle. Identification of modules with a high risk of containing defects allows development teams and project managers to prioritize the allocation of testing resources, to optimize resource utilization, and to minimize the quantity of quality assurance effort that is expended unnecessarily. The outputs of the defect prediction algorithm are used as input to support decision-making for software process management.

The proposed approach is evaluated experimentally utilizing datasets of software defects collected from NASA, which are representative of industrial software projects. The experimental results show that reasonable prediction performance can be obtained, especially when evaluating large and complex datasets. The results of the study provide evidence that defect prediction analytics is a viable method for improving software development processes through proactive quality assurance, use of metrics to make decisions, and improved predictability of project outcomes.