Unsupervised Learning based Backlog Prioritization in Distributed Agile Software Development

Abstract

Distributed Agile software development has attracted the consideration of recent industrial fields. Agile technologies rely on faster delivery of shippable product to the customer which was a limitation of earlier traditional technologies. Due to faster delivery requirements, the project should be developed well within its time frame. This may be possible only if backlogs are properly handled and resolved which includes certain new features, changes to existing features or other tasks as identified by any of the stakeholders.  At the end of each sprint an ordered list of remaining backlogs should be ready with the project managers for their review and allotment to the team members. The ordering has to be done based upon stakeholders’ priorities based business challenges, task dependencies, available skill set and budget limitations. In this way, proper and timely backlog prioritization becomes an important activity of Agile based software development. This activity becomes more critical in distributed Agile environment. In this paper, we have proposed and implemented supervised learning based prediction technique for backlog prioritization. Based on available task set (in the form of user stories) and allotted team member’s previous experience, the model predicts probable backlogs.  The list of probable backlogs (most appropriate among existing user stories), is passed to the project managers and team members. If these probable backlogs are available initially, team members pay more emphasis on those user stories, to finish them timely. Thus, the technique acts as a backlog prevention mechanism. A comparison has been made based on the total number of available backlogs after the end of each sprint. The results show that the proposed model gives better results in terms of backlog prediction and reduction in number of unattended user stories.