Mining developer communication data streams

TL;DR

This paper applies data stream mining techniques to developer communication data from a software repository to identify metrics that predict build outcomes, demonstrating the significance of a few key metrics.

Contribution

It introduces the use of Hoeffding Tree and ADWIN methods to analyze developer communication streams for predicting build success or failure.

Findings

Few communication metrics significantly predict build outcomes

Hoeffding Tree effectively classifies build results

Concept drift detection improves prediction accuracy

Abstract

This paper explores the concepts of modelling a software development project as a process that results in the creation of a continuous stream of data. In terms of the Jazz repository used in this research, one aspect of that stream of data would be developer communication. Such data can be used to create an evolving social network characterized by a range of metrics. This paper presents the application of data stream mining techniques to identify the most useful metrics for predicting build outcomes. Results are presented from applying the Hoeffding Tree classification method used in conjunction with the Adaptive Sliding Window (ADWIN) method for detecting concept drift. The results indicate that only a small number of the available metrics considered have any significance for predicting the outcome of a build.