Aprendizagem de máquina para análise de indicadores em processos de software

• Ano de defesa: 2015
• Autor(a) principal: Bodo, Leandro [UNESP]
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Estadual Paulista (Unesp)
• Palavras-chave em Português: Computer science; Ciência da computação; Engenharia de software; Engenharia de requisitos; Software - Controle de qualidade; Desempenho - Medição; Aprendizado do computador
• Link de acesso: http://hdl.handle.net/11449/154707; http://www.athena.biblioteca.unesp.br/exlibris/bd/cathedra/01-08-2017/000869005.pdf

Software development companies have been facing problems and challenges in relation to software quality for decades. Quality management involves three basic processes: quality planning, quality assurance and quality control. The quality control process provides information to evaluate the performance and changes in projects, processes or products. For this, performance indicators should be defined and analyzed in order to help decision-making. During the monitoring of the software production processes, data of the performance indicators is collected and stored in historical bases in order to be analyzed by the managers of the processes. Statistical quality control techniques aid the evaluation of the collected data. However, some aspects complicate the appropriate analysis to timely decision-making. The quality control may require analyzing groups of indicators, composed of indicators of various processes and with different granularity, types and collect frequency. Furthermore, as the volume of indicator data increases, the complexity of analysis also tends to increase. In this context, this work presents a systematic for analysis of performance indicators, using semi-supervised machine learning techniques.This systematic consists of steps that cover the selection of indicators, the process of labeling and the analysis of data collected in monitoring. In addition, the work presents a reference model to support the selection of indicators, considering the processes of the levels G and F of the MPS model for software (MPS-SW). The other developed reference model is based on the perspectives of the Balanced Scorecard model, in order to support the definition of groups of indicators. The labeling process and the data analysis process are done in a single step. Information visualization techniques are used to support the labeling process. The work presents a case study on the systematic presented, using real data from a software development ...