DBSCAN parameter explorer

What this project does

This small project demonstrates how to use DBSCAN (Density-Based Spatial Clustering of Applications with Noise) to discover irregular groups in two-dimensional data and to flag isolated points as noise. You will generate a synthetic 2D dataset with clusters of differing density and some isolated points, then fit DBSCAN across a small grid of epsilon and minimum-neighbour settings. The code reports cluster sizes, the fraction of noise, and computes silhouette scores when there are at least two non-noise clusters. A concise parameter explorer output helps you understand how density settings change group assignments.

Data

The dataset is synthetic and built in-code using sklearn.datasets.make_blobs plus a set of isolated points added explicitly. This creates a small two-dimensional dataset suitable for visualising density-based clustering behaviour: some dense clusters, some sparser clusters, and clear isolated points that DBSCAN should label as noise depending on parameters.

How to build it

Create a Python 3.11+ environment with the exact requirements listed in requirements.txt.
Save project.py and test_project.py from this package into a folder.
Install dependencies: pip install -r requirements.txt
Run the project or tests as described below.

How to run

Run the module directly: python project.py

This will print a JSON object describing the best parameter setting found (by silhouette score), the corresponding number of clusters, the overall noise fraction and a small explorer listing results for each tested (eps, min_samples) pair.

How to test

Run the pytest suite: pytest -q

The tests check that the run function returns the expected keys, that numeric metrics lie in reasonable ranges, that the main held-out metric meets a conservative floor, and that the function is deterministic for a fixed seed.

Results

Measured by running python project.py (seed 42) on 2026-10-23; the tests in test_project.py assert the floor.

Metric	Value
median_pairwise_distance	5.0365
best_silhouette	0.7153
best_eps	1.1000
best_min_samples	3
best_n_clusters	4
best_noise_fraction	0.0235

Also returned (see python project.py output): explorer (12 items).

Limitations

This project uses synthetic data built into the script; it is not a substitute for domain-specific datasets. DBSCAN sensitivity to scale and the choice of epsilon means a single global epsilon may not work for datasets with very uneven local densities. Silhouette score is provided only when there are at least two non-noise clusters; it is not defined for a single cluster or when all points are noise.

Next steps

Extend the explorer to include more fine-grained epsilon values and automated suggestions (e.g. k-distance plots).
Add small visual checks (scatter plots) when running locally to inspect cluster shapes and noise.
Experiment with preprocessing (scaling) when features have different units or ranges.

Full case study, build steps and evaluation checks: DBSCAN clustering implementation on networkershome.com

Built by Networkers Home · Learn the skills in the AI Engineer Course · Part of 100 AI projects with code