for my “value calendar” reader. I would like to thank you for your support.
This Google sheet has been around for years. They appear little by little. But when it's time to publish them, I always need hours to rearrange everything, clean up the layout, and make them fun to read.
Today, we are going to DBSCAN.
DBSCAN does not learn a parametric model
Just in lof, DBSCAN -I parametric model. There is no formula to keep, no rules, no centimeter, and nothing to reuse later.
We have to keep All data Because the size structure depends on every point.
Its full name Density-based spatial management of noise applications.
But be careful: This “personality” is not a Gaussian equation.
It is based on support the idea of humanity. “How many neighbors live near me”.
Why DBSCAN is special
As its name suggests, Dbscan does two things at the same time:
- It finds clusters
- Marks Anomalies (points that are not clustered)
This is why I present the algorithms in this way:
- K– Pillars and Hmm there are Integration models. They released the glossary: k-means measurements, means and variances of GMM.
- Forest of Separation and The filter there are Anomaly detection models. Their only goal is to get unusual points.
- DBSCAN stays in the middle. It does both Anomaly detection and detectionbased on the idea of the personality of the neighbors.
Small dataset to keep things accurate
We stay with the same small data we used for LOF: 1, 2, 3, 7, 8, 12
If you look at these numbers, you already see two clear groups:
one around 1-2-3one around 7-8again What you recorded Living alone.
DBSCAN captures these concepts well.
Summary in 3 steps
DBSCAN asks Three simple questions For each point:
- How many neighbors do you have within a small radius (eps)?
- Do you have enough neighbors to be the main point (minpts)?
- Now that we know the basic points, which group do you belong to?
Here is a summary of the albscan algorithm in 3 steps:
Let's start step by step.
Dbscan in 3 steps
Now that we understand the concept of personality and neighborhood, DBSCAN becomes much easier to explain.
Every algorithm is compatible with it Three Easy Steps.
Step 1 – Count the neighbors
The goal is to check how many neighbors each point has.
We take a small radius called EPS.
For each point, we look at all the other points and mark those that are farthest from the EPS.
These are the ones neighbors.
This gives us a first impression of the difficulty:
A point with many neighbors is in a dense region,
A point with few neighbors resides in a high-level region.
For an example of a 1-three tone like ours, a common choice is:
EPS = 2
We draw a small circle of radius 2 around each point.
Why is it called EPS?
Name EPS it comes from a Greek book ε (epsilon)traditionally used in mathematics to represent a plural or a small radius around the point.
So in DBSCAN, EPS it's actually the “smallest neighbor radius”.
It answers the question:
How far do we look at each point?
So in Excel, the first step is to combine i Battery Plate Matrixthen count how many neighbors each point has within EPSs.
Step 2 – High points and density connection
Now that we know the neighbors from step 1, we are working you've stolen it to determine which points Context.
Mintpts mean here the minimum amount of points.
It is the smallest number of neighbors the point must (within the APS Radius) to consider a Context the point.
A point is basic if at least four you've stolen it Neighbors inside EPS.
Otherwise, it might be The border or An argument.
and EPS = 2 and mintpts = 2we have 12 non-core.
When the maximum score is unknown, we simply check what the score is Mass-accessible from them. If a point can be reached by moving from one point to another within the EPS, it belongs to that group.
In Excel, we can represent this as a simple connection table that shows which points are connected to which neighbors are at a higher level.
This connection is what Dbscan uses to create the Clusters in Step 3.
Step 3 – Assign labels to groups
The goal is to convert communications into real collections.
When the connection matrix is ready, clusters appear naturally.
DBSCAN groups all connected points together.
To give each group a simple and generated name, we use a very precise rule:
A Cluster label is the smallest node in a connected group.
For example:
- The group {1, 2, 3} becomes a cluster 1
- The group {7, 8} becomes cluster 7
- The point is like What you recorded Apart from high-class neighbors eba An argument
This is exactly what we will demonstrate in Excel using formulas.
Final thoughts
Dbscan is perfect for teaching a local perspective.
No probability, no gaussian formula, no measurement step.
Just distances, neighbors, and a little radio.
But this is simple and limiting.
Because DBSCAN uses a single set radius for everyone, it cannot adapt to situations where the dataset contains different clusters.
HDBSCAN keeps the same intuition, but looks just Radii also keeps that constant.
It's very strong, and very close to how they naturally see clusters.
