A Bagging-Based Ensemble Approach for Enhanced 3D Shape Clustering

Raydonal Ospina (UFBA)

Abstract: This study proposes the use of the Bagging procedure to enhance the stability and consistency of three-dimensional shape clustering. In Statistical Shape Analysis, morphological variability and the requirement for geometric alignment (translation, rotation, and scale adjustments) pose significant challenges to the consistency of clustering algorithms. To mitigate these effects, bootstrap resampling is employed to generate multiple training sets, thereby reducing sensitivity to variations in the data. The method was integrated with the K-means, CLARANS, and Hill Climbing algorithms. The results demonstrate that the Bagging-based approach improves performance and efficiency in the classification of complex geometric structures, such as those found in neuroimaging data, compared to single-run clustering methods.