Unsupervised learning in ML

Introduction

Unsupervised learning is a type of machine learning where a computer learns from data that does not have predefined labels or answers. Instead of being told what the correct result is, the algorithm studies the data and identifies patterns, groups, or relationships on its own.

Unsupervised learning is useful when there is a large amount of data but no labelled information available. It helps organizations discover hidden patterns and gain a better understanding of their data.

The two major areas of unsupervised learning discussed in this article are:

1. Clustering
2. Dimensionality Reduction

1.CLUSTERING

Clustering is an unsupervised learning technique that groups similar data points together. Data points within the same group, known as a cluster, are more similar to each other than to data points in other groups.

Uses of Clustering

Clustering can be used for:

• Customer segmentation – grouping customers according to their behaviour or characteristics.
• Market research – identifying different groups within a market.
• Image analysis – grouping similar images or parts of images.
• Fraud detection – identifying unusual patterns in transactions.
• Document classification – grouping documents with similar topics.
• Recommendation systems – identifying users with similar interests.
• Healthcare analysis – grouping patients with similar characteristics or patterns.

HIERARCHICAL CLUSTERING

Hierarchical clustering is a clustering technique that creates a hierarchy of groups. Instead of simply dividing data into a fixed number of clusters, it shows how individual data points and groups are related at different levels.

There are two main approaches to hierarchical clustering:

I).Agglomerative Clustering

Agglomerative clustering is a bottom-up approach.

It starts with each data point as its own individual cluster. The algorithm then repeatedly combines the most similar clusters until all the data points form one large cluster or the desired grouping is reached.

Process:

Individual Data Points → Small Clusters → Larger Clusters → One Large Cluster

Agglomerative clustering is the more commonly used approach to hierarchical clustering.

II).Divisive Clustering

Divisive clustering is a top-down approach.

It begins with all data points in one large cluster. The algorithm then repeatedly divides the cluster into smaller groups until individual groups are formed or the desired number of clusters is reached.

Process:

One Large Cluster → Smaller Clusters → Individual Groups

Summary

Approach	Direction	Starting Point
Agglomerative	Bottom-up	Individual data points
Divisive	Top-down	One large cluster

---

DENDROGRAM

A dendrogram is a tree-like diagram used to display the results of hierarchical clustering.

It shows how individual data points or groups are combined or divided at different levels.

The height at which two groups are connected represents how different or distant they are. By choosing a particular level on the dendrogram, we can determine how many clusters to create.

Example of a simple dendrogram

              ┌───────────────┐
              │               │
          ┌───┴───┐       ┌───┴───┐
          │       │       │       │
          A       B       C       D

Here, A and B are grouped together, while C and D form another group.

Importance of a Dendrogram

Dendrograms help users:

• Understand how data points are related.
• Visualize hierarchical clusters.
• Decide how many clusters should be created.
• Identify groups with similar characteristics.

DBSCAN

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

It is a clustering algorithm that groups data points based on how closely packed they are. It can identify areas with many data points while treating points that do not belong to a dense group as noise or outliers.

Unlike some clustering methods, DBSCAN does not require the user to specify the exact number of clusters beforehand.

How DBSCAN Works

DBSCAN looks for areas where data points are close together.

It generally identifies three types of points:

• Core points – points located within a sufficiently dense area.
• Border points – points near a dense cluster but not dense enough to be core points.
• Noise points – points that do not belong to any cluster.

Advantages of DBSCAN

• Can find clusters with different shapes.
• Can identify outliers and noise.
• Does not require the number of clusters to be specified beforehand.
• Useful for spatial and location-based data.

Disadvantages of DBSCAN

• Its performance can be affected when clusters have very different densities.
• Choosing suitable parameter values can be difficult.
• It may not work well with very high-dimensional data without appropriate preparation.

2.DIMENSIONALITY REDUCTION

Dimensionality reduction is the process of reducing the number of variables or features in a dataset while trying to preserve the most important information.

Dimensionality reduction transforms the data into fewer dimensions while attempting to retain its important patterns.

Benefits of Dimensionality Reduction

• Makes large datasets easier to analyze.
• Reduces computational requirements.
• Can make data easier to visualize.
• Helps remove unnecessary or redundant information.
• Can improve the performance of some machine learning algorithms.

PRINCIPAL COMPONENT ANALYSIS (PCA)

Principal Component Analysis (PCA) is one of the most widely used dimensionality reduction techniques.

PCA transforms a dataset with many variables into a smaller number of new variables called principal components.

The principal components are created so that they capture as much of the important variation in the original data as possible.

Uses of PCA

PCA can be used for:

• Reducing the number of features.
• Visualizing high-dimensional data.
• Removing redundant information.
• Simplifying datasets before further analysis.
• Improving computational efficiency.

Advantages of PCA

• Reduces the number of variables.
• Makes complex datasets easier to visualize.
• Can reduce redundancy in data.
• Can make some machine learning processes faster.

Disadvantages of PCA

• The new components can be difficult to interpret.
• Some information may be lost during the reduction.
• The results can be affected by the scale of the variables.
• It may not be suitable when every original variable has important meaning.

APPLICATIONS OF UNSUPERVISED LEARNING

1.Marketing

Businesses can use clustering to divide customers into groups based on their purchasing habits, interests, or behaviour. This helps businesses understand different customer groups.

2.Banking and Finance

Unsupervised learning can help identify unusual transaction patterns and discover groups of customers with similar financial behaviour.

3.Healthcare

Healthcare organizations can analyze patient data to identify groups with similar characteristics or discover patterns in medical information.

4.Retail

Retail businesses can analyze purchasing behaviour to identify groups of customers and discover which products are commonly purchased together.

5.Image Processing

Clustering can be used to group similar pixels or identify patterns within images. Dimensionality reduction can also help simplify complex image data.

6.Cybersecurity

Unsupervised learning can identify unusual patterns in network activity that may require further investigation.

7.Social Media

Organizations can analyze large amounts of social media information to identify groups, topics, trends, or patterns in user behavior.

8.Manufacturing

Manufacturers can use unsupervised learning to identify unusual machine behaviour and discover patterns that may indicate equipment problems.