Combinando Planning Poker e aprendizado de máquina para estimar esforço de software

Abstract

Estimating software effort is a critical factor to organizations, as underestimation or overestimation can result in project failures. Planning Poker is one of the most used practices to define effort estimates. The estimate is based on the experience of the members, through a meeting involving all team members. However, the information generated in this debate is not kept due to the informality of the practice and how that knowledge is lost, there is no way to use it in future estimates. The application of machine learning (ML) techniques in effort estimates has grown in recent years, used in addition to or alternatively to other approaches. Studies indicate that the use of combined practices provides greater assertiveness in relation to individual techniques. Therefore, this study aimed to describe the combination of Planning Poker with AM, an approach created and named ML Planning Poker, and assess whether the proposal interferes with the effort estimation process. We carry out bibliographic research, a systematic mapping and a survey strengthening the study bases. Based on the findings, we describe the steps of ML Planning Poker and we developed a tool that supports and an interactive medium with the teams in the estimation process, such that built into it we create an ML model. After that, we document the proposal and, we evaluate with undergraduate students and IT professionals. The evaluation with students resulted in identical assertiveness in the tasks estimated using the original Planning Poker and the proposal. Although, regarding tasks with incorrect estimation, we noticed that ML Planning Poker had a better result, given that, 57.1% of the tasks had a difference of at most 1 hour between estimated and performed time compared to 39.2% of the original Planning Poker. Furthermore, of the participating students, 81.2% agree that the proposal contributes to the estimation process. IT professionals realized benefits of the proposal and defended that ML provides a subsidy to members. They also reinforced the problem of forgetting very old tasks, making the current estimate difficult, and ML Planning Poker helps, as it brings similar tasks from the previous ones. Even though there are issues to be improved, such as the accuracy of the model and the usability of the tool, insights point to the benefits of ML Planning Poker in tasks that have been performed for a long time or that members have no experience, bringing greater security to the participant in the definition of his estimate. ML Planning Poker has potential because the human factor present in Planning Poker continues to be considered while ML supports decision making, allowing to improve the estimation process.