Análise comparativa de métodos de clusterização em dados geoespaciais

Abstract

This study presents a comparative analysis of clustering methods applied to geospatial data from municipalities in Rio Grande do Sul, focusing on the performance and representativeness of the resulting groups. The evaluated methods include K-means, Weighted K-means, Agglomerative Clustering, and DBSCAN, using geographic and population data. The methodology incorporated fi xed centroids, inspired by Aline Gomes's thesis, to align automatic methods with regional contexts. Results showed that K-means and Weighted K-means outperformed in terms of cluster cohesion and separation, particularly in densely populated regions. Conversely, DBSCAN was eff ective in detecting outliers but faced challenges when adapted to fi xed centroids. The analysis emphasized that integrating automatic methods with manual adjustments can off er more eff ective hybrid solutions, combining mathematical effi ciency with qualitative sensitivity. This study contributes to enhancing clustering techniques in geospatial analyses, providing valuable insights for public policy and regional planning.