DBScan Clustering

• Why needed
  • K-means Clustering does bad when the clusters are nested
• Overlap Region: A predefined radius for a circular region
• Core Points: If a point has K points in their overlap region, then those points are called core points.
• As it is ordered, the ordering change the cluster
  • Like for a non-core point if it has 2 cluster in their overlap region, it will be assigned to the first one.

Steps:

1. Find all core points
2. Randomly assign a centroid from the core points
   1. Assign all the points in the overlap region of that point to that cluster
   2. Run Step 2.1 for all the newly assigned points
3. For each not-core points check if it has a core point in the overlap region assigned to a cluster
   1. If yes, then assign that not-core point to that cluster
4. If there is more core points, go to step 2