Hidden Markov Models for the Prediction of Developer Involvement Dynamics and Workload

Abstract

The evolution of software projects is driven by developers who are in control of the developed artifacts. When analyzing the behavior of developers, the observable behaviors are, e.g., commits, messages, or bug assignments. For defining dynamic activities and workload of developers, we consider underlying characteristics, which means the level of involvement according to their role in the project. In this paper, we propose to employ Hidden Markov Models (HMMs) to model this underlying behavior given the observable behavior as input. For this, we observe monthly commits, bugfixes, mailing list activity, and bug comments for each developer over the project duration. As output we get a model for each developer describing how likely it is to be in a low, medium, or high contribution state of every point in time. As a result, we discovered that same developer types exhibit similar models in terms of state patterns and transition matrices, which represent their involvement dynamics. Although the workload of the different developer roles related to this is more complex to model, we created a general model which performs nearly as well as individual developer contribution models. Moreover, to demonstrate the practical applicability, we present an example of the usage of our approach in project planning.