Unsupervised Learning and Clustering Techniques

Unsupervised Learning

Analyzing data without labeled response variables.

Supervised Learning

Predicting outcomes using features and response variables.

Clustering

Grouping similar observations into distinct subgroups.

Hidden Structures

Unseen patterns within the data set.

Distance Measure

Metric to quantify similarity between data points.

Euclidean Distance

Straight-line distance between two points in space.

Manhattan Distance

Distance measured along axes at right angles.

Cosine Similarity

Cosine of the angle between two vectors.

High-dimensional Space

Data represented by vectors of large size.

Subgroups Discovery

Identifying distinct groups within a dataset.

Data Visualization

Representing data to reveal patterns or insights.

Subjectivity in Analysis

Interpretation varies based on data context.

Gene Expression Measurements

Data used to group breast cancer patients.

Shoppers Characterization

Grouping shoppers by browsing and purchase history.

Movie Ratings Clustering

Grouping movies based on viewer ratings.

Clustering Problem Setup

Grouping points based on proximity in data.

Similarity Definition

Criteria for determining observation closeness.

Sky Objects Catalog

2 billion objects characterized by 7 radiation dimensions.

Cluster Members

Observations within a cluster are similar.

Dissimilar Clusters

Members of different clusters are not alike.

Domain-specific Consideration

Knowledge-based criteria for similarity assessment.

Broad Class of Methods

Various techniques for subgroup discovery in data.

Cosine Similarity

Ranges from -1 (opposite) to 1 (same)

Orthogonality

Indicates zero similarity between vectors

Cosine Distance

Calculated as 1 minus cosine similarity

Sparse Data

Data with many zero values, often binary

Jaccard Similarity

Measures similarity between finite sample sets

High Distance

Indicates low similarity between points

Low Distance

Indicates high similarity between points

Hierarchical Clustering

Clusters formed through agglomerative or divisive methods

Agglomerative Clustering

Combines nearest clusters into one cluster

Divisive Clustering

Starts with one cluster and splits recursively

K-means Clustering

Partitions data into K distinct, non-overlapping clusters

Centroid

Average of all data points in a cluster

Euclidean Distance

Assumed distance metric in K-means clustering

Initial Cluster Assignment

Randomly assign points to clusters or select centroids

Convergence in K-means

No points move between clusters, centroids stabilize

Cluster Assignment Process

Assign points to nearest centroid iteratively

Random Initialization Effect

Random selection can lead to different clustering results

Compact Clusters

Clusters with smallest distances within themselves

Selecting K

Determining the optimal number of clusters

Iterative Process

Reassign points and update centroids until stable

K-means Algorithm

Iteratively assigns points to clusters based on centroids

Distance Metric

Measures similarity or dissimilarity between data points

K-means clustering

Requires pre-specifying the number of clusters K.

Silhouette score

Measures cluster cohesion versus separation.

Silhouette range

Values range from -1 to +1.

High silhouette value

Indicates good cluster matching.

Hierarchical clustering

Does not require pre-defined number of clusters.

Agglomerative clustering

Bottom-up approach to cluster merging.

Dendrogram

Visual representation of hierarchical clustering.

Centroid

Average location of points in a cluster.

Cluster merging

Repeatedly combine nearest clusters until stopping criterion.

Euclidean distance

Distance measure for determining cluster nearness.

Key operation

Combine two nearest clusters iteratively.

Stopping criterion

Condition to end the clustering process.

Cluster representation

Location of clusters determined by centroids.

Distance measurement

Assessing cluster proximity using centroid distances.

Data point

Individual observation in a clustering dataset.

Centroid example

Average of data points in a cluster.

Fusion height

Indicates similarity of merged observations.

Bottom of dendrogram

Indicates high similarity between observations.

Top of dendrogram

Indicates low similarity between observations.

Preventing feature dominance

Solutions to avoid skewed clustering results.

Cluster analysis

Examining data points grouped into clusters.

Picking k

Choosing the optimal number of clusters.

Cohesion

Similarity of an object to its own cluster.

Feature Scaling

Rescaling features to a common range, e.g., [0,1].

Euclidean Distance

A common measure of similarity in clustering.

Manhattan Distance

Distance calculated as the sum of absolute differences.

Cosine Similarity

Measure of similarity based on angle between vectors.

Jaccard Index

Similarity measure for comparing sets of data.

Pearson Correlation

Statistical measure of linear correlation between variables.

Clustering

Grouping data points based on similarity.

Cluster Centroid

Average point representing a cluster's members.

Exploratory Analysis

Analyzing data to discover patterns without prior hypotheses.

Cluster Profiling

Describing and understanding characteristics of clusters.

Fresh Food Lovers

Cluster of customers favoring organic and fresh foods.

Distance Measure Choice

Selecting appropriate metric for clustering data analysis.

Data Preparation

Preprocessing data before applying clustering algorithms.

Customer Segmentation

Dividing customers into groups for targeted marketing.

Transaction History

Record of customer purchases used for clustering.

Dendrogram

Tree-like diagram representing data clustering hierarchy.

Unsupervised Learning

Learning patterns from data without labeled responses.

Customer Features

Attributes like age and income used for clustering.

Marketing Strategies

Tailored approaches based on customer segment characteristics.

Cluster Analysis

Technique to identify patterns in data without explanations.

Buying Behavior

Patterns in customer purchases used for segmentation.

Data Mining

Extracting useful information from large datasets.

Numerical Features

Quantitative attributes used for analysis and clustering.