Home

Services

About Us

Projects

Contact Us

Disclaimer

Unsupervised clustering using nearest neighbor likelihood 

Nearest neighbor likelihood values are computed using a critical distance threshold. These nearest neighbor likelihood values are then used to identify level sets that are used to find data points that are far from points that have previously been clustered. In this context 'far' means outside the critical distance threshold. The first cluster is assumed to contain the most likely data point, which is used to start the algorithm, and approximate cluster centers are assigned as the most likely data point in each cluster. Singleton clusters and isolated clusters containing a small number of points are frequently found. These may be removed or assigned to the closest cluster.

Description	Data	Clusters
A one-dimensional example using a mixture of two normal distributions. The plots show the nearest neighbor likelihood values on the y-axis for the data and the identified clusters.
A two-dimensional example using a mixture of five normal distributions.
A two dimensional example using a data set consisting of  concentric rings. These data were first converted to magnitude values by computing their distances from the origin and then clustered in one-dimension.
A two-dimensional example with structured data consisting of four spiral arms.
A three dimensional example using a mixture of five normal distributions.