Um sistema de apoio à decisão para priorização e estruturação de histórias de usuários: um suporte para equipe ágeis

Abstract

Despite the advances obtained in the Software Engineering with agile methods the market still shows successful rates in unsatisfactory projects. Throughout a software project, changing requirement and priority are unavoidable and recurrent. This is an important factor to the software development businesses mainly small ones with limited resources. The usage of agile techniques such as Scrum and User Stories (US) benefits the businesses and make them more competitive. One of the problems faced with agile requirements is to obtain a prioritization of secure US according to the business value given by the client in agreement with technical specifications. This paper proposes to present an essay that serves as basis to the construction of a decision support tool to the US prioritization decisions. For such was considered a software project as a temporary organization and suggested metrics that best suits the needs of small teams avoiding reworks, the increasing in the deadlines and costs, without disregarding the client satisfaction in interactive deliveries. Based on the Volere requirements template and in the literature review it was proposed the adoption of five metrics to be considered in the prioritization of the User Stories: the client satisfaction in receiving the US and the client dissatisfaction in not receiving it; replacing the traditional value of business; the so usual complexity; the necessity of team learning and; software risks. Using online forms were captured data about these metrics and the prioritization given to each US in software projects that uses the agile Scrum method. Decision Tree was the proposed suggestion for predicting the US prioritization for having a practical visualization and a more intuitive interpretation, facilitating the acceptance as a decision method support for professionals involved in the area. In spite of the low databases volume the results obtained through the Weka tool, like the precision and the ROC curve, were satisfactory without prediction tendencies and with good indexes of correctness, after the algorithms and databases adjustments avoiding overfitting and underfitting.