Task purpose estimation in software development based on automatic measurement data and machine learning

In this paper we propose a method to support Personal Software Process (PSP), which is a well known software process improvement framework for individual developers. The proposed method estimates developer's purposes (aims) from time-series data about developer's tasks, given by an execution history of software applications. We implemented the method by a machine learning algorithm, Random Forests. The experiment result shows the prediction with the time-series data is more accurate than the prediction without the time-series data. Especially, when using longer time-series data, accuracy of estimation became 97 %. It can be expected that the proposed method can help developers' process improvement as they become aware of how much time they spent on a specific aim such as implementation and testing.