SMILE: uma proposta de método para gestão de projetos de aprendizado de máquina
Carregando...
Data
2022-06-23
Autores
Fukuda, Mauricio Makoto
Título da Revista
ISSN da Revista
Título de Volume
Editor
Universidade Estadual Paulista (Unesp)
Resumo
A utilização de algoritmos de aprendizado de máquina tem apresentado um aumento expressivo ao longo dos últimos anos devido à redução dos custos e ao aumento da capacidade de processamento dos computadores. O estudo do aprendizado de máquina é importante devido ao potencial dessa tecnologia, demonstrado por diversas aplicações, desde o controle de qualidade na manufatura de bens de consumo, até o diagnóstico de doenças. Essas aplicações promovem benefícios tais como a redução de custos operacionais e o aumento da precisão de diagnósticos. Neste contexto, o aprofundamento da discussão sobre os métodos para gestão de projetos específicos para o aprendizado de máquina se mostra relevante e se justifica devido à escassez de estudos sobre esse tema na literatura e pelo potencial de resposta às críticas relativas à utilização adaptada de métodos antigos, como o CRISP-DM, que não foi originalmente desenvolvido para projetos de aprendizado de máquina. Esta pesquisa teve como objetivo desenvolver um método para gestão de projetos de aprendizado de máquina denominado SMILE, acrônimo de Structured Machine Learning Employment, que combina elementos de métodos tradicionais, ágeis e híbridos, para gestão de projetos. Esse método busca abordar os desafios específicos da gestão de projetos de aprendizado de máquina, identificados por meio de revisão da literatura e entrevistas com gestores de projetos, analistas e cientistas de dados que desenvolvem produtos e serviços utilizando o aprendizado de máquina. A proposta do método SMILE foi desenvolvida utilizando-se o Design Science como método de pesquisa e foi apresentada neste trabalho em formato de procedimento com o detalhamento de cada uma de suas etapas, incluindo a ideação, decisão de investimento, coleta de dados, desenvolvimento e implantação, de forma a facilitar o entendimento sobre sua aplicação. O método SMILE foi apresentado para especialistas que avaliaram a sua aplicabilidade. Tais especialistas foram os mesmos entrevistados na etapa da pesquisa de levantamento dos desafios específicos para os projetos de aprendizado de máquina. Eles avaliaram o método de maneira positiva, o que demonstra que o método proposto tem potencial para superar os desafios e lacunas existentes em projetos dessa natureza.
The use of machine learning algorithms has shown a significant increase over the last few years due to the reduction of costs and increase in the processing capacity of computers. The study of machine learning is essential due to the potential of this technology, demonstrated by the various applications, from quality control in the manufacture of consumer goods to the diagnosis of diseases. These applications provide benefits such as reduced costs and increased diagnostic accuracy. In this context, the deepening of the discussion on specific project management methods for machine learning is relevant and justifiable due to the scarcity of studies on this topic in the literature and its potential to respond to criticisms related to the adapted use of old methods, such as CRISP-DM, which was not originally developed for machine learning projects. This research aimed to develop a machine learning project management method called SMILE, an acronym for Structured Machine Learning Employment, which combines elements of traditional, agile, and hybrid project management methods. This method seeks to address the specific challenges of machine learning project management. Challenges and gaps in managing machine learning projects were identified through literature review and interviews with project managers, analysts, and data scientists who develop products and services using machine learning. The proposal of the SMILE was developed using Design Science as a research method and was presented in this work in procedural format with the detailing of each of its stages, including ideation, investment decision, data collection, development, and implementation, in order to facilitate the understanding of its application. The SMILE method was presented to experts who evaluated its applicability. The experts were the same ones interviewed in the research stage of surveying the specific challenges for machine learning projects. Respondents evaluated the method positively regarding its applicability, which demonstrates that the proposed method has the potential to address the challenges and gaps existing in projects of this nature.
The use of machine learning algorithms has shown a significant increase over the last few years due to the reduction of costs and increase in the processing capacity of computers. The study of machine learning is essential due to the potential of this technology, demonstrated by the various applications, from quality control in the manufacture of consumer goods to the diagnosis of diseases. These applications provide benefits such as reduced costs and increased diagnostic accuracy. In this context, the deepening of the discussion on specific project management methods for machine learning is relevant and justifiable due to the scarcity of studies on this topic in the literature and its potential to respond to criticisms related to the adapted use of old methods, such as CRISP-DM, which was not originally developed for machine learning projects. This research aimed to develop a machine learning project management method called SMILE, an acronym for Structured Machine Learning Employment, which combines elements of traditional, agile, and hybrid project management methods. This method seeks to address the specific challenges of machine learning project management. Challenges and gaps in managing machine learning projects were identified through literature review and interviews with project managers, analysts, and data scientists who develop products and services using machine learning. The proposal of the SMILE was developed using Design Science as a research method and was presented in this work in procedural format with the detailing of each of its stages, including ideation, investment decision, data collection, development, and implementation, in order to facilitate the understanding of its application. The SMILE method was presented to experts who evaluated its applicability. The experts were the same ones interviewed in the research stage of surveying the specific challenges for machine learning projects. Respondents evaluated the method positively regarding its applicability, which demonstrates that the proposed method has the potential to address the challenges and gaps existing in projects of this nature.
Descrição
Palavras-chave
Aprendizado de máquina, Métodos ágeis, Gerenciamento de projetos, Inteligência artificial, Machine learning, Agile methods, Project management, Agile-stage-gate, Artificial intelligence, Administração de projetos, Tecnologia de ponta