Un enfoque de machine learning para apoyar la identificación de la deuda técnica en arquitectura

Abstract

Decisions made by architects to favor short-term objectives and to the possible detriment of long-term software quality are known as Architectural Technical Debt. This type of technical debt is difficult to identify because it is related to quality attributes not visible to the customer, such as the maintainability and evolvability of the system. Objective: For this reason, in this paper, a supervised machine learning model is proposed to support the identification of technical debt in architecture that is located in the design stage of the architecture. Methods: This proposal relies on the information collected from the artifacts produced during the architecture design to build a dataset to evaluate different supervised learning algorithms and thus establish the one that offers the best accuracy. The identification of the technical debt, within the framework of this proposal and unlike those proposed in the literature, does not consider the source code. Results: The performance of the model was evaluated through a real industry case and allowed discovering that both accuracy and recall present acceptable levels. Conclusions: The data used to train the model, while appropriate, has room for further improvement. This approach will allow architects to support the identification of conscious and unconscious ATD injected into their architectures.