In this lesson, we used geometric data sets to compare the performance of the clustering algorithms k-means, k-medoids, and DBSCAN. Our three examples, each with a visualization and Mathematica code to show the clustering results, formed the framework for this.
Example 1: Two hundred well spaced random points fill a lemniscate.
DBSCAN takes these points and finds 4 clusters.
The clusters are displayed using the k-means approach with k=2 and k=4.
Similarly, k=2 and k=4 are used to illustrate the k-medoids approach.
Example 2: A union of a circle and an annulus filled with 400 randomly placed points is used to repeat the experiment.
Within this data set, DBSCAN detects two clusters.
Using k=2 and k=4, the k-means and k-medoids approaches are once more used.
Example 3: In this example, the area of a square is filled in less than a
400 random spots in the maximum circle.
Four clusters are found in this scenario by DBSCAN.
Using k=2 and k=4, the k-means and k-medoids approaches are illustrated.
comparison study and a brief description of each technique in the next release