Bagging Voronoi classifiers for clustering spatial functional data


Statistical learning
Bagging Voronoi classifiers for clustering spatial functional data
Saturday 2nd July 2011
Secchi, P.; Vantini, S.; Vitelli, V.
Download link:
We propose a bagging strategy based on random Voronoi tessellations for the exploration of high dimensional spatial data, suitable for different purposes (e.g., classification, regression, ...). In particular, we consider the problem of clustering functional data indexed by the sites of a spatial finite lattice. The analysis is based on local representatives from neighboring data, i.e., belonging to the same element of a tessellation: the proposed algorithm accounts for spatial dependence by repeatedly clustering functional local representatives with respect to a random system of neighborhoods. Due to the resampling of tessellations, classification result is a cluster assignment frequency map, which can be used to define an a-posteriori criterion to choose the most suitable grouping structure. Thanks to spatial dependence, local representatives are expected to be less noisy and less correlated than original data, providing better performances. Moreover, this reduction in the dimension of the dataset permits the handling of high dimensional sets of data otherwise intractable without an explicit model for spatial dependence. The performance of the proposed approach is tested on simulated data. An application to environmental data contained in Surface Solar Energy database is also illustrated.
This report, or a modified version of it, has been also submitted to, or published on
International Journal of Applied Earth Observation and Geoinformation