Utilizando Algoritmos de Machine Learning Caixa Branca para Avaliar a Manutenibilidade dos Modelos de Features

• Autor(a) principal: Silva, Publio Blenilio Tavares
• Data de Publicação: 2020
• Tipo de documento: Trabalho de conclusão de curso
• Idioma: por
• Título da fonte: Repositório Institucional da Universidade Federal do Ceará (UFC)
• dARK ID: ark:/83112/00130000109q7
• Texto Completo: http://www.repositorio.ufc.br/handle/riufc/58831
• topic: Aprendizado do Computador; Engenharia de Linha de Produto de Software; Manutenibilidade; Modelo de Características

Software Product Lines (SPL) are generally specified using a Feature Model (FM), an artifact designed in the early stages of the SPL life cycle. This artifact can easily become too complex, which makes it difficult to maintain SPL. Therefore, it is essential to continuously assess the maintainability of this artifact, which can be done through measurements. Some approaches that assess the maintainability of the FM through the aggregation of maintenance measures are described in the literature. Machine Learning (ML) models can be used to create these approaches, as they can aggregate the values of a set of independent variables into a single target data, also called a dependent variable. Besides, if white box ML models are used, it is possible to interpret and explain the results obtained with this type of ML model. In this context, this work proposes to create ML white-box models to classify the maintainability of the FM, based on 15 maintenance measures. To build the models, we performed the following steps: (i) we compared two approaches to assess the maintainability of FMs through an oracle of FM maintainability classifications made manually by 15 SPL specialists; (ii) we use the best of the two approaches to pre-classify the ML dataset training dataset; (iii) we generated 3 ML models and compared them with the metrics of classification accuracy, precision, recall, F1 and AUC-ROC; and, (iv) we use the best of the 3 models to create a mechanism capable of providing improvement indicators to domain engineers. The best of the 3 models used the decision tree algorithm that obtained accuracy, precision, and recall of 0.81, F1 of 0.79, and AUC-ROC of 0.91. Using this model, we were able to reduce the number of measures needed to assess the maintainability of the FM from 15 to 9 measures. The decision tree generated by the algorithm was used to create a mechanism capable of providing suggestions for changing maintenance measures so that the maintainability of the artifact improves.