Implementación web de algoritmos de clustering con restricciones de tamaño

Abstract

Size-constrained clustering is a technique in the field of machine learning that seeks to group data by ensuring that each group meets specific predefined size criteria. This approach helps provide an equal or unequal distribution of data within particular groups, which is essential for specific applications. In this study, two clustering algorithms, SC-Medoids, and CSCLP, were implemented in a web application developed with R Studio, the Shiny package, and its deployment of the service in Shinyapps.io. The methodological process adopted a waterfall model, with sequential phases that helped to structure the development of the website, thus generating an application that allows users to perform clustering in real-time and adjust parameters interactively. Stress, load, execution time, and RAM usage tests were performed to validate the stability and efficiency of the system under certain conditions. In addition, surveys evaluated usability, which provided insights into user experience, addressing aspects such as ease of use, intuitive navigation, and overall satisfaction level. The results show a high level of acceptance and recommendation of the web application. After performing post-deployment tests, it has been verified that the page supports up to 50 active users simultaneously. In contrast, the application can process approximately 90 requests in about five seconds in stress tests. As for the processing of the SC-Medoids algorithm, it supports datasets of up to 3500 instances, compared to CSCLP, which supports up to 1800.