Um processo baseado em redes bayesianas para avaliação da aplicação do scrum em projetos de software.

Abstract

The use of Agile Software Development (ASD) is increasing to satisfy the need to respond to fast moving market demand and gain market share. In contrast with traditional plan-driven processes, ASD are people and communication-oriented, flexible, fast, lightweight, responsive, driven for learning and continuous improvement. As consequence, subjective factors such as collaboration, communication and self-management are key to evaluate the maturity of agile adoption. Scrum, which is focused on project management, is the most popular agile method. Whenever adopted, the usage of Scrum must be continuously improved by complementing it with development and management practices and processes. Even though the Retrospective Meeting, a Scrum event, is a period at the end of each sprint for the team to assess the development method, there are no clear and specific procedures to conduct it. In literature, there are several, but no consolidated, proposed solutions to assist on ASD adoption and assessment. Therefore, the research problem is: how to instrument Scrum to assist on the continuous improvement of the development method focusing on the requirements engineering process, development team and product increment? In this thesis, we propose a Bayesian networks-based process to assist on the assessment of Scrum-based projects, instrumenting the software development method to assist on its continuous improvement focusing on the requirements engineering process, development team and product increments. We have built the Bayesian network using a Knowledge Engineering Bayesian Network (KEBN) process that calculates the customer satisfaction given factors of the software development method. To evaluate its prediction accuracy, we have collected data from 18 industry projects from one organization through a questionnaire. As a result, the prediction was correct for fourteen projects (78% accuracy). Therefore, we conclude that the model is capable of accurately predicting the customer satisfaction and is useful to assist on decision-support on Scrum projects.