UNSUPERVISED

LEARNING
PRESENTED BY,
DVEIN INNOVATIONS
 K-MEANS
CLUSTERING
 INTRODUCTION

 K-Means is an unsupervised clustering algorithm that groups

similar data points into K distinct clusters.

 It finds natural groupings in data based on feature similarity

— without any labels.
 When to Use K-Means?

Use Case Example Goal

Customer segmentation Group similar customers
Market basket analysis Cluster similar shopping
patterns
Image compression Reduce colors into clusters
 WORKING

Step 1: Choose the Number of Clusters (K)

 You specify the number of clusters you want (e.g., K = 3)

 Choosing K is user-defined or based on methods like the Elbow

Method

Step 2: Initialize K Centroids Randomly

 Pick K random points as the initial centroids (center of the clusters)

Step 3: Assign Each Data Point to the Nearest Centroid

 Use Euclidean distance to measure closeness

 Each data point is assigned to the cluster of the nearest centroid

Step 4: Recalculate the Centroids

 For each cluster, calculate the mean of all points in the cluster

 This new mean becomes the new centroid

Step 5: Repeat Until Convergence

 Repeat Steps 3 and 4 until:

 Centroids don’t change (or change very little)

 A maximum number of iterations is reached

 KEY CONCEPTS

Term Meaning
Cluster A group of similar data points
Centroid The center point of a cluster
Inertia The sum of distances from points
to their centroid
Convergence When centroids no longer move
Elbow Method A technique to choose the best K
 CHARACTERISTICS OF K-MEANS

Feature Description
Type Unsupervised
Input Raw data (no labels)
Output Cluster assignment for each point
Assumes Clusters are spherical and equal in
size
Works well when Clear, well-separated clusters exist
 PRO’S & CON’S

ADVANTAGES LIMITATIONS
 Simple and Easy to Implement  Sensitive to initial centroid positions

 Fast and Efficient on Large Datasets  Requires pre-defined K

 Scalable to Big Data  Not good for non-spherical clusters

 Works Well with Clearly Separable Clusters  Affected by outliers

 Easy to Interpret and Visualize in 2D/3D  Sensitive to feature scaling

HIERARCHICAL CLUSTERING
 INTRODUCTION

 Hierarchical Clustering is an unsupervised learning

algorithm that builds a hierarchy of clusters either from the
bottom up or top down.

 It produces a tree-like structure (dendrogram) showing how data

points are merged or split at different levels.
 When to Use Hierarchical Clustering?

Use Case Example Goal

Gene expression analysis Find relationships among genes
Customer grouping Understand nested customer
types
Social network analysis Detect communities or groups
 WORKING

Two Main Types:

 Agglomerative (Bottom-Up) – Most common

 Divisive (Top-Down) – Less common

Step 1: Treat Each Data Point as a Cluster

 Start with each point as its own cluster

 So, if there are 100 points → we begin with 100 clusters

Step 2: Compute Distance Between All Clusters

 Use a distance metric (usually Euclidean)

 Also choose a linkage method:

 Single Linkage – min distance between any 2 points

 Complete Linkage – max distance between any 2 points

 Average Linkage – average distance

 Ward’s Method – minimizes variance

Step 3: Merge the Closest Clusters

 Merge the two closest clusters into one

 Now we have one fewer cluster

Step 4: Repeat Until All Points Are in One Cluster

 Continue merging until there’s only one big cluster

 This creates a hierarchical tree (dendrogram)

Step 5: Choose the Number of Clusters

 Use the dendrogram and cut it at a desired height

 This height determines the number of clusters

 KEY CONCEPTS

Term Meaning
Dendrogram A tree diagram showing the hierarchy
Linkage Method to compute distance between
clusters
Agglomerative Start with points, merge upward
Divisive Start with one big cluster, split
downward
Cutting the dendrogram Selects number of final clusters
 PRO’S & CON’S

ADVANTAGES LIMITATIONS
 No Need to Specify Number of  Not scalable

Clusters (K)  No backtracking

 Produces a Tree-like Structure  Sensitive to noise

(Dendrogram)  Choice of linkage matters
 Easy to Visualize Relationships

 Flexible Distance and Linkage

Options
DBSCAN
 INTRODUCTION

 DBSCAN is an unsupervised clustering algorithm that groups

data points that are closely packed together (high density areas)
and identifies outliers as noise.

 Unlike K-Means, DBSCAN doesn’t require predefining the

number of clusters and can detect clusters of arbitrary shapes.
 When to Use DBSCAN?

Use Case Example Goal

Geospatial data analysis Cluster locations
Anomaly detection Identify outliers
Image segmentation Detect dense regions
 WORKING

Two Important Parameters

 Eps (ε) – The maximum radius to consider around a data point.

 MinPts – Minimum number of points required to form a dense

region.
 Step 1: Classify Each Point

 For each data point, check how many neighbors it has within the
radius ε.

 Based on this, points are classified into:

Type of Point Condition

Core Point Has at least MinPts within ε
Border Point Has fewer than MinPts within ε but is in the
neighborhood of a Core Poin
Noise Point Neither Core nor Border
Step 2: Form Clusters

 Start with a random point.

 If it’s a Core Point, form a new cluster with its neighbors.

 Expand the cluster by visiting all neighbors’ neighbors recursively.

Step 3: Mark Noise

 Points that don’t fit into any cluster are labeled as noise or outliers.

Step 4: Repeat

 Repeat the process until all points are visited and clustered or
marked as noise.
 CHARACTERISTICS OF DBSCAN

Feature Description
Doesn’t require K No need to predefine clusters
Handles outliers Marks them explicitly
Can detect complex shapes Unlike K-Means
Distance metric Typically Euclidean
 PRO’S & CON’S

ADVANTAGES LIMITATIONS

 No Need to Specify Number of  Bad choices can lead to poor

Clusters (K) results

  Struggles when clusters have

Identifies Outliers Naturally
different densities
 Handles Noise Well
 Distance becomes less
meaningful
APRIORI ALGORITHM
 INTRODUCTION

 Apriori is an unsupervised learning algorithm used

for association rule mining to find frequent itemsets in
transactional datasets and to uncover associations between items.

 It is widely used in Market Basket Analysis — finding patterns like

"customers who buy X also buy Y."
 WHEN TO USE APRIORI?

Use Case Example Goal

Market basket analysis Which products are bought
together
Recommendation systems Suggest complementary
products
Inventory management Stock frequently bought items
together
 WORKING
Step 1: Set Thresholds
 Support (s): Minimum frequency an itemset must have to be
considered.
 Confidence (c): Minimum likelihood of the rule being true.
 Lift: How much more likely the consequent is given the antecedent.
Step 2: Generate Frequent Itemsets
 Start with individual items → count their support (frequency).
 Prune items that don't meet the minimum support threshold.
 Example:
 {Milk}, {Bread}, {Butter} → check support
 Remove those with low support.
 Step 3: Build Larger Itemsets

 Combine frequent items to form pairs, triplets, etc.

 At each step, eliminate combinations that contain infrequent

subsets (Apriori Principle).

 Example:

 From {Milk}, {Bread} → create {Milk, Bread} if both are frequent.

Step 4: Generate Association Rules
 Once frequent itemsets are identified:
 For each frequent itemset, generate rules like:
 A → B (If A is bought, B is likely to be bought)
 Calculate:
 Confidence: Probability of B given A.
 Lift: Strength of the rule compared to random chance.

Step 5: Filter Rules Based on Confidence & Lift

 Only rules that exceed minimum confidence and lift thresholds are
selected as strong rules.
 PRO’S & CON’S

ADVANTAGES LIMITATIONS
 Easy to Understand and Implement  Computationally expensive

  Not scalable for large datasets

Provides Clear Association Rules

  Needs proper thresholds

Works Well for Market Basket

Analysis  Only finds frequent patterns

 No Need for Labeled Data

THANK YOU!