Results & Implementation

This section details the performance of our custom DBSCAN implementation and visualizes the results on various datasets.

Algorithm Logic

Implementation follows the standard density-based clustering approach:

1. For each point, we find all neighbors within eps radius using Euclidean distance.

2. If a point has at least min_samples neighbors, it becomes a Core Point and starts a new cluster.

3. Using a queue visit all density-reachable neighbors to expand the cluster.

4. Points that are not reachable from any Core Point are labeled as -1 (Noise).

Visual Results

We tested the algorithm on synthetic datasets that are notoriously difficult for distance-based algorithms like K-Means.

Moons Dataset

The “Moons” dataset consists of two interleaving half-circles. Standard K-Means would fail here by drawing a straight line through the middle. DBSCAN successfully follows the shape.

Circles Dataset

The “Circles” dataset contains a smaller circle inside a larger one. This is a classic topology problem.