Clustering approaches and ensembles applied in the delineation of management classes in precision agriculture.

Abstract. This paper describes an experiment performed using different approaches for spatial data clustering, aiming to assist the delineation of management classes in Precision Agriculture (PA). These approaches were established from the partitional clustering algorithm Fuzzy c-Means (FCM), traditionally used in this context, and from the hierarchical clustering algorithm HACCSpatial, especially designed for this PA task. We also performed experiments using traditional ensembles approaches from the literature, evaluating their behavior to achieve consensus solutions from individual clusterings obtained from features splitting or running one of the abovementioned algorithms. Results showed some differences between FCM and HACC-Spatial, mainly for the visualization of management classes in the form of maps. Considering the consensus clusterings provided by ensembles, it became clear the attempt to achieve an agreement result that most closely matches the original clusterings, showing us some details that may go undetected when we analyse only the individual clusterings.