Estimativa do Tempo de Resolução de Issues no GitHub Usando Atributos Textuais e Temporais

Resumo

Estimating issues resolution time is one of the most important steps in software maintenance processes. However, although the subject is covered in the literature, there are few specific models for GitHub. This platform is very popular mainly in the open source context but its issue tracking system is not bureaucratic and issues are registered in a very simple way, which makes the process of building predictive models even more challenging. This work aims to develop machine learning models to estimate the resolution time of issues from GitHub. To handle the data scarcity, we propose textual attributes to capture issues characteristics; and temporal attributes to provide information about the time of issue events. Neural networks were used in classification algorithms and proved to be more suitable for solving this problem. To validate the proposed models we compared them with a reference from literature through different metrics and the results were positive with a significant improvement in accuracy.