Skip to content

README.md

Latest commit

 

History

History
105 lines (73 loc) · 3.05 KB

README.md

File metadata and controls

105 lines (73 loc) · 3.05 KB

DBSCAN Clustering from Scratch (No ML Libraries, No Classes)

This notebook walks you through a simple implementation of DBSCAN (Density-Based Spatial Clustering of Applications with Noise) using only NumPy. It is designed for students and beginners to understand how clustering by density works, without relying on scikit-learn.

What You’ll Learn

  • What DBSCAN is and how it works
  • Key terms: core points, border points, and noise
  • How to implement DBSCAN step-by-step:
    • Compute point-to-point distances
    • Find dense neighborhoods (region queries)
    • Expand clusters recursively
  • How to visualize DBSCAN results

What is DBSCAN?

DBSCAN groups together points that are close to each other (high-density regions), and marks points in low-density regions as outliers (noise).

Main Parameters:

  • eps: The distance threshold to search for neighboring points
  • min_samples: Minimum number of neighbors (including self) to form a dense cluster

Point Types:

Type Definition
Core Point Has at least min_samples neighbors within eps
Border Point Has fewer than min_samples, but is within eps of a core point
Noise Neither core nor border (considered outlier)

Files Included

File Description
dbscan_from_scratch.ipynb Full DBSCAN implementation with visualization
README.md This file describes the purpose and usage

How to Use

  1. Clone or download this repository
  2. Open the notebook dbscan_from_scratch.ipynb in:
    • Jupyter Notebook, OR
    • Google Colab
  3. Run each cell in order:
    • Generate sample clustered data with outliers
    • Implement DBSCAN logic step-by-step
    • Visualize the discovered clusters and noise

Requirements

This project requires:

  • Python 3.x
  • NumPy
  • Matplotlib

Install dependencies:

pip install numpy matplotlib

Best Use-Cases for DBSCAN

  • Clustering data with irregular shapes
  • Detecting anomalies or outliers
  • Geospatial clustering (e.g., identifying areas of interest)
  • When the number of clusters is not known ahead of time
  • When you’re okay with labeling some points as noise and don’t want to force every point into a cluster.

Why DBSCAN?

  • No need to specify the number of clusters (k)
  • Automatically detects outliers (noise)
  • Works well for non-spherical clusters (unlike K-Means)

Suggested Experiments

  • Try different eps and min_samples values
  • Compare with K-Means to observe differences
  • Visualize step-by-step region expansion
  • Apply to real-world datasets (e.g., GPS coordinates, customer behavior)

License

This project is open-source and licensed under the MIT License.

Contribute

Contributions are welcome; feel free to optimize the implementation, visualize regions, or add interactive widgets for parameter tuning.