DBSCAN Algorithm of Unsupervised Machine Learning

1️. Introduction to DBSCAN

DBSCAN stands for Density-Based Spatial Clustering of Applications with Noise.

It is an unsupervised machine learning clustering algorithm used to group data points based on density (i.e., how closely packed data points are within a specific region) rather than distance from a centroid (like K-Means).

DBSCAN is especially powerful for:

✔ Detecting clusters of arbitrary shapes (clusters that are not necessarily circular or uniform in structure)
✔ Identifying outliers (noise)
✔ Works well with real-world noisy data (noisy data means data containing errors, random variations, irrelevant values, or extreme outliers that do not belong to any meaningful pattern)

---

2️. Why DBSCAN is Important?

Unlike traditional clustering algorithms such as K-Means, DBSCAN:

✔ Does NOT require the number of clusters in advance
✔ Can detect non-linear cluster shapes
✔ Automatically detects noise
✔ Performs well when data contains irregular distributions

---

3️. Core Concept Behind DBSCAN

DBSCAN is based on one simple idea:

"Clusters are dense regions of data points separated by low-density regions."

Here, density means the number of data points present within a defined neighborhood (radius) around a particular point.

It uses two important parameters:

1. Epsilon (ε)

• Radius of neighborhood around a data point.
• Defines how close points must be to be considered neighbors.

2. MinPts (Minimum Points)

• Minimum number of points required inside ε-radius to form a dense region.

---

4️. Key Terminologies in DBSCAN

There are 3 types of points:

1. Core Point

A point that has at least MinPts points within ε radius.

2. Border Point

A point that has fewer than MinPts but lies within ε of a core point.

3. Noise (Outlier)

A point that is neither core nor border.

 

5️. Visual Understanding of DBSCAN

From the images above, you can observe:

·         Clusters are not circular

·         Some points are labeled as noise

·         Arbitrary-shaped clusters (clusters that can take any irregular form instead of fixed shapes like circles) are formed

6️. How DBSCAN Works (Step-by-Step)

Step 1:

Select an unvisited point.

Step 2:

Find all points within ε distance.

Step 3:

If neighbors ≥ MinPts → mark as Core Point and start cluster.

Step 4:

Expand cluster recursively by checking neighbors of neighbors.

Step 5:

If neighbors < MinPts → mark as Noise (temporarily).

Step 6:

Repeat until all points are processed.

7️.  Logic Behind the Algorithm

The main logic is:

🔹 Density Reachability

A point A is density reachable from B if:

·         B is a core point

·         A lies within ε of B

🔹 Density Connectivity

Two points are density connected if there exists a chain of density-reachable points connecting them.

This recursive connectivity builds clusters.

8. Distance Metrics Used in DBSCAN

DBSCAN commonly uses:

·         Euclidean Distance

·         Manhattan Distance

·         Minkowski Distance

·         Haversine Distance (for geo data)

Distance selection depends on:

·         Nature of data

·         Feature scaling

·         Dimensionality

 

9. When to Use DBSCAN?

Use DBSCAN when:

✔ You don’t know number of clusters
✔ Clusters are irregular in shape
✔ Data contains noise/outliers
✔ Working with spatial/geographical data
✔ Fraud detection
✔ Anomaly detection

Avoid DBSCAN when:

·         Dataset is very high dimensional

·         Varying density clusters exist 

Real-Life Applications of DBSCAN

📍 1. Geographical Data Clustering

·         Grouping houses in a city

·         Earthquake hotspot detection

·         Traffic analysis

📊 2. Fraud Detection

·         Detect abnormal banking transactions

📡 3. Image Processing

·         Object segmentation

🧬 4. Medical Data

·         Gene clustering

·         Disease pattern detection

🚗 5. Autonomous Vehicles

·         Obstacle detection

Problem Statement :  (for DBSCAN on Iris dataset):

To apply the DBSCAN clustering algorithm on the Iris flower dataset using Sepal and Petal measurements in order to identify natural density-based clusters without predefining the number of clusters.

To evaluate whether DBSCAN can effectively distinguish different Iris species and detect any noise or overlapping patterns based on feature similarity.

Download Dataset - Iris Dataset

 Dataset Description

• The dataset contains measurements of Iris flowers.

• It includes four numerical features: Sepal Length, Sepal Width, Petal Length, and Petal Width.

• All measurements are recorded in centimeters.

• Each row in the dataset represents one individual flower sample.

• The dataset also includes one categorical variable called Species.

• The Species variable has three classes: setosa, versicolor, and virginica.

• The numerical features describe the morphological (physical) characteristics of the flowers.

• The dataset is widely used for clustering and classification analysis.

• It helps analyze how different flower species can be grouped based on similarity in their measurements.

 Go to Jupyter Notebook to use DBSCAN Algorithm