A Geospatial Implementation of a Novel Delineation Clustering Algorithm Employing the K-means

The overarching objective of this paper is to introduce a novel Fast, Efficient, and Scalable k-

means

k-means (

FES-k

-

means*

) algorithm. This algorithm is designed to increase the overall performance of the standard k-

means

clustering technique. The

FES-k

-

means*

algorithm uses a hybrid approach that comprises the k-d tree data structure, the nearest neighbor query, the standard k-

means

algorithm, and Mashor’s adaptation rate. The algorithm is tested using two real datasets and two synthetic datasets and is employed twice on all four datasets. The first trial consisted of previously

MIL-SOM*

trained data, and the second was on raw, untrained data. The approach presented with this method enables unfounded knowledge discovery, otherwise unclaimed by conventional clustering methods. When used in conjunction with the

MIL-SOM*

training technique, the

FES-k

-

means

* algorithm reduces the computation time and produces quality clusters. In particular, the robust

FES-k-means*

method opens doors to (1)

faster

cluster production than conventional clustering methods, (2)

scalability

allowing application in other platforms, and its ability to handle small and large datasets, compact or scattered, and (3)

efficient

geospatial data analysis of large datasets. All of the above makes

FES-k-means*

live up to defending its well-deserved name—Fast, Efficient, and Scalable

k-means

(

FES-k-means*

). The findings of this study are vital to the relatively new and expanding subfield of geospatial data management.