Online platform for building, testing and deploying predictive models

• Autor(a) principal: Infante, Gonçalo Brito
• Data de Publicação: 2017
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10362/45651
• topic: Machine learning; Predictive models; Online platform; Flexible system; Supervised-learning; Classification; Domínio/Área Científica::Engenharia e Tecnologia::Engenharia Eletrotécnica, Eletrónica e Informática

Machine Learning (ML) and Artificial Intelligence (AI) have been traditionally built and deployed manually in a single machine, using tools such as R or Weka. Times are changing and in the real-time service and big data era, this methods are being obsoleted, as they severely limit the applicability and deployability of ML. Many companies such as Microsoft, Amazon and Google have been trying to mitigate this problem developing their MLaaS (Machine Learning as a Service) solutions, which are online platforms capable to scale and automate the development of predictive models. Despite the existence of some ML platforms available in the cloud, that enable the user to develop and deploy ML processes, they are not suitable for rapidly prototype and deploy predictive models, as some complex steps need to be done before the user starts using them, like configuration of environments, configuration of accounts and the overcome of the steep learning curve. In this research project, it’s presented MLINO, which is a concept of an online platform that allows the user to rapidly prototype and deploy basic ML processes, in an intuitive and easy way. Even though the implementation of the prototype wasn’t the optimal, due to software and infrastructure limitations, through a series of experiments it was demonstrated that the final performance of the prototype was satisfactory. When benchmarking the devised solution against the Microsoft Azure ML, the results showed that MLINO tool is easier to use, and takes less time when building and deploying a basic predictive model.