DreamLearning: Um site para o estudo e aplicação de aprendizado de máquina

Abstract

This paper presents the development of a learning platform, called DreamLearning, aimed at teaching and applying machine learning algorithms. This proposal seeks to fill existing gaps in the teaching of this area for undergraduate students, in order to implement an accessible, interactive tool, based on open source software, in the Portuguese language. To this end, the adopted methodology followed principles of software engineering, including the clear definition and modeling of requirements, as well as the use of open technologies, such as Scikit-learn, Flask and Bootstrap. As a result, the developed platform allows the performance of classification, regression and associative analysis tasks, using supervised and unsupervised algorithms. Among the functionalities available in the educational environment, we highlight the training and testing of models, hyperparameter adjustments, cross-validation, visualization of performance metrics and, finally, the export of the generated source code. Additionally, a conversational agent was implemented with the objective of clarifying students’ doubts during the learning process. The platform also offers video lessons, developed based on pedagogical methodologies, seeking to provide a better understanding of the content, in addition to graphical visualizations and access to the history of the simulations performed. Usability tests were conducted with undergraduate students, where it was observed that DreamLearning is a promising tool for teaching machine learning algorithms. The proposed platform, in an accessible and interactive way, introduces students to fundamental concepts of the area, allowing practical simulation and access to the code of the algorithms used.