Avaliação e melhoria contínua do trabalho em equipe no desenvolvimento ágil de software com Redes Bayesianas.

Abstract

Agile Software Development adoption has increased due to its characteristics that allow fast responses to business requirements change in the volatile software development indus- try. According to the agile principles and to studies published in the literature, teamwork related factors have positive impact on the success of ASD projects. In the literature, there are some instruments with the purpose of assessing Teamwork Quality, but they might not be useful in concrete situations that demand prediction, and problem diagnosis or progno- sis. Therefore, in this thesis, we present a process based on a Bayesian Network model that enables measuring, prediction, assessment, and continuous improvement of Teamwork Quality in the context of Agile Software Development. The model was built by following a top-down approach based on the knowledge of a specialist, and it can adapted to fit a given context. The proposed solution, comprised by the model and process, was evaluated in re- gards to its practical utility in a case study with three teams of one organization, and the results were positive. Given the fact that the model presented has a questionnaire as source of data, we evaluated the level of agreement between respondents in different roles within a team, including project manager, technical leader, and development teams. As a result, we verified that the results with considerable level of agreement were calculated only for the the managers and development teams roles. Moreover, we realized an empirical study with two organizations to compare the model presented in thesis with another model presented in the literature, in regards to their composition, including their variables and measures. According to the results of the study, considering only the set of equivalent variables presented in both models, they are similar from the perspective of their constructs; and the results calculated by both models, in general, are also similar. Finally, we present practical use cases based on real experiences with the purpose of exemplifying how to use the model presented in this thesis.