Gerenciador de datasets com visualização gráfica de meta-features utilizando o MFE

Abstract

With the concern of extracting the best solution in machine learning algorithms, several studies raised the question of how to identify which are the characterization alternatives that can reveal different relevant results within a dataset. To solve this problem, the MFE, a meta-features extraction software, can be used, but its use implies knowledge in programming language, making it complex and requiring a high learning curve through trial and error. The objective of this work is to build a tool capable of removing this dependence on the user and making its use easier and more intuitive. For this, a system composed of a Web page was developed, for visualization and manipulation of datasets and a server, applying clean architecture techniques, for handling user requests and processing data through the MFE. As a result, software was obtained with a simple interface that requires little instruction for its use, which allows managing ARFF type datasets and provides a comparative graphical view from the results obtained from the MFE. The application of micro services architecture and SOLID principles and clean architecture during software development made it an easy maintenance and extension solution. To improve the software, it was suggested for future work the implementation of a functionality that allows the selection of the type of data extraction and also the use of technologies such as Docker and Kubernetes to improve the robustness of the system.