From past to future: An experience using data mining to guide tests.

Abstract

It’s common to face errors during the process of software development. Be it an agile or traditional methodology, those errors are documented and registered in tools that allow us to manage and trace them. This data is rich in information about the product we are developing and the processes being used. Therefore, the analysis of this data can give us a better view of the product’s characteristics, its faults and how they affect it’s quality. Having said that, this article relates the use of Machine Learning techniques in a software’s error data base, to identify and classify critical areas in the system, in order to support decision making from the test team, the evolution process and production code maintenance by the developers. Overall, a set of 1045 software defects registries were collected, and we could identify that: (i) 63% of the defects are concentraded in 10 of the 71 existing functionalities, (ii) a functionality has a tendecy to show defects in the last versions of our software, (iii) the software have 4 critical functionalites that concentrate 52% of the reported defects and show recurrent defects.