Ferramenta para representação gráfica do algoritmo de aprendizado de máquina k-Nearest Neighbors

Abstract

The growing advancement of Artificial Intelligence and, in particular, Machine Learning, has dri- ven the application of algorithms across various sectors. However, the inherent complexity of many of these algorithms, especially in high-dimensional scenarios, represents a challenge for their understanding. This work focuses on this challenge by proposing the development of an interactive web tool for the visualization of the k-Nearest Neighbors algorithm. To build the solution, a modular architecture based on design patterns was adopted, and a custom graphics library was developed. Additionally, the t-Distributed Stochastic Neighbor Embedding algorithm was integrated to enable dimensionality reduction. The system was validated through exploratory and end-to-end testing, ensuring compliance with the requirements, along with static analyses to assess code quality. As a result, the application allows users to submit their own datasets, ad- just essential hyperparameters of the algorithm, and visualize the resulting decision boundaries directly in the browser, confirming the delivery of a functional and portable solution.