Machine Learning Methods and Tools

Explain the method and why you think it’s suitable for your use case. Explain the choice of tools/packages/data and the reason for use.

Machine Learning algorithm - DBSCAN

This project will use the Density-based spatial clustering of applications with noise (DBSCAN) algorithm [Schubert et al., 2017], which is a clustering technique well-suited for the point cloud dataset we’ll be working with. DBSCAN is able to group together points that are densely clustered without setting the expected number of clusters a priori (as in k-means clustering). Specifically, we will be using cuML’s implementation of DBSCAN that is GPU-accelerated to find dense clusters across millions of points in minutes.

Animation sourced from https://dashee87.github.io/data science/general/Clustering-with-Scikit-with-GIFs. See also https://www.naftaliharris.com/blog/visualizing-dbscan-clustering for a nice interactive way of visualizing the DBSCAN algorithm and how the parameters affect the clustering result.