A Study on Architectural Smells Prediction

Abstract

Architectural smells can be detrimental to the system maintainability, evolvability and represent a source of architectural debt. Thus, it is very important to be able to understand how they evolved in the past and to predict their future evolution. In this paper, we evaluate if the existence of architectural smells in the past versions of a project can be used to predict their presence in the future. We analyzed four Java projects in 295 Github releases and we applied for the prediction four different supervised learning models in a repeated cross-validation setting. We found that historical architectural smell information can be used to predict the presence of architectural smells in the future. Hence, practitioners should carefully monitor the evolution of architectural smells and take preventative actions to avoid introducing them and stave off their progressive growth.