Title: Connecting historical changes for cross-version software defect prediction

Authors: Xue Bai; Hua Zhou; Hongji Yang; Dong Wang

Addresses: School of Software, Yunnan University, Kunming, Yunnan, China ' College of Big Data and Intelligent Engineering, Southwest Forestry University, Kunming, Yunnan, China ' School of Informatics, Leicester University, Leicester, England, UK ' School of Software, Yunnan University, Kunming, Yunnan, China

Abstract: In the whole software life cycle, software defects are inevitable and increase the cost of software development and evolution. Cross-Version Software Defect Prediction (CVSDP) aims at learning the defect patterns from the historical data of previous software versions to distinguish buggy software modules from clean ones. In CVSDP, metrics are intrinsic properties associated with the external manifestation of defects. However, traditional software defect measures ignore the sequential information of changes during software evolution process which may play a crucial role in CVSDP. Therefore, researchers tried to connect traditional metrics across versions as a new kind of evolution metrics. This study proposes a new way to connect historical sequence of metrics based on change sequence named HCSM and designs a novel deep learning algorithm GDNN as a classifier to process it. Compared to the traditional metrics approaches and other relevant approaches, the proposed approach fits in projects with stable and orderly defect control trend.

Keywords: software testing; cross-version defect prediction; software metrics; historical change sequences; deep learning; DNN; deep neural networks; gate recurrent unit.

DOI: 10.1504/IJCAT.2020.10032601

International Journal of Computer Applications in Technology, 2020 Vol.63 No.4, pp.371 - 383

Received: 29 Jun 2020
Accepted: 29 Jun 2020

Published online: 19 Oct 2020 *

Full-text access for editors Access for subscribers Purchase this article Comment on this article