Many clustering algorithms, like K-Means, assume that clusters are roughly spherical. But what if your data isn't so neatly organised? What if your clusters are long and winding, or have irregular shapes? This is where density-based clustering shines.

DBSCAN, which stands for Density-Based Spatial Clustering of Applications with Noise, is an algorithm that groups points based on how closely packed they are. It excels at finding non-spherical clusters and is also great at identifying outliers, which it treats as noise.

Instead of partitioning the entire dataset, DBSCAN starts at an arbitrary point and begins to grow a cluster by finding its dense neighbours. The process continues until it can no longer find any more nearby points to add. This approach allows clusters to grow into any shape, as long as the points remain densely connected.

To define what 'dense' means, DBSCAN relies on two key parameters:

1. Epsilon (ε): This is a distance value. It defines a radius around each data point, creating a neighbourhood.
2. Minimum Points (MinPts): This is a number. It sets the minimum count of points that must fall within a point's ε-neighbourhood for it to be considered part of a dense region.

Together, these two parameters set the threshold for density. An area is considered 'dense' if a point has at least MinPts within its ε radius.

Based on the ε and MinPts settings, DBSCAN classifies every single point in the dataset into one of three types. This classification determines how clusters are formed.

Next, we have border points.

Finally, there are the points that don't belong anywhere.

Let's visualise how these definitions work. Imagine we set MinPts to 4. We can see how points P, Q, and N are classified based on their neighbours.

The clustering process itself is straightforward and builds on these point classifications:

1. The algorithm picks an arbitrary, unvisited point.
2. It checks if the point is a core point. If not, it's temporarily marked as noise (it could become a border point later).
3. If it is a core point, a new cluster is created. The algorithm finds all points that are density-reachable from this core point and adds them to the new cluster. This includes other core points and any border points on the cluster's edge.
4. This process repeats until all points have been visited and classified.

A key concept here is 'density-reachable'. A point p is density-reachable from a point q if there's a chain of core points leading from q to p. This is how DBSCAN connects points to form clusters of arbitrary shapes.

This method is powerful because it doesn't force every point into a cluster. By having a dedicated category for noise, DBSCAN provides a more realistic view of datasets that contain outliers or aren't perfectly structured.

Now that you understand the core principles of DBSCAN, you're ready to see how it's applied in practice.