Abordagem heurística baseada em busca em vizinhança variável para o agrupamento balanceado de dados pelo critério da soma mínima das distâncias quadráticas

Abstract

After advances in collecting and storing data and the growth in applications that provide new information, the number of data elements available is huge in both volume and variety. With this increase in the quantity of information, the need to understand them and summarize them has become increasingly urgent. The Balanced Clustering seeks to find groups of similar entities that have approximately the same size. In this dissertation, we propose a new heuristic approach based on metaheuristic Variable Neighborhood Search (VNS) and methodology "Less is More Approach"(LIMA) to data clustering problem using the criterion of the minimum sum-of-squared distances applying balancing restriction for the groups. The algorithms found in the literature are not scalable, while the problem of increased size in addition to elements 5000 in accordance with experiments performed in this study. The computational experiments show that the proposed method outperforms the current state of the art for the problem.