Exploring the Intersection Between Software Maintenance and Machine Learning - A Systematic Mapping Study

While some areas of software engineering knowledge present great advances with respect to the automation of processes, tools and practices, areas like software maintenance have scarcely been addressed either by industry or academia, delegating the solution of technical tasks or human capital to manual or semiautomatic forms. In this context, machine learning (ML) techniques play an important role when it comes to improving maintenance processes and automation practices that can accelerate delegated but highly critical stages when the software launches. The aim of this article is to gain a global understanding of the state of ML-based software maintenance using the compilation, classification and analysis of a set of studies related to the topic. The study was conducted by applying a systematic mapping study protocol, characterized by the use of a set of stages that strengthen replicability. The review identified a total of 3,776 research articles, subject to four filtering stages, ultimately selecting 81 articles that were analyzed thematically. The results reveal an abundance of proposals that use neural networks applied to preventive maintenance and case studies that incorporate ML in subjects of maintenance management and management of the people who carry out these tasks. In the same way, a significant number of studies lack the minimum characteristics of replicability.