Machine Learning Class: Clustering and Applications

Overview

• Topics Covered: Clustering, applications of clustering, clustering as a machine learning task, example of identifying a professor for a machine learning subject, types of clusters.

Clustering

Definition

• Purpose: Finding subgroups or clusters in a dataset based on object characteristics.
• Process: Raw data (unlabeled) -> Clustering algorithm -> Subgroups or clusters.
• Effectiveness: Objects in the same cluster are similar; objects from different clusters are different.
• Labeling: After clustering, each group or cluster gets a label.

Example: Movie Promotion

• Use Case: Advertisement company wants to target movie promotions to specific groups in a country.
• Data: Includes age, location, financial condition, political stability.
• Clustering Analysis: Based on browsing patterns, likes/dislikes, frequently visited sites.
• Outcome: Produces targeted promotions to specific groups (e.g., youngsters for sports movies).

Applications of Clustering

Text Data Mining

• Tasks: Text categorization, text clustering, document summarization, concept extraction, sentiment analysis, entity relation model.

Customer Segmentation

• Parameters: Demographics, financial conditions, buying habits, likes/dislikes.
• Usage: Retailers/advertisers promote products to specific customer segments.

Anomaly Checking

• Use Cases: Fraud detection in bank transactions, unauthorized computer intrusions, suspicious movements on radar.

Data Mining

• Purpose: Extracting required data from large datasets (e.g., terabyte-size datasets).
• Process: Grouping features from extremely large datasets.

Clustering as a Machine Learning Model

• Purpose: Discovering data patterns instead of predicting single outputs
• Nature: Unsupervised machine learning task that divides data into clusters/groups.
• Process: Analyzes unlabeled data to identify internal patterns and groups data based on characteristics.
• Output: Produces new data and labels objects with class labels based on analyzed data patterns.
• Effectiveness: Measured by homogeneity within groups and heterogeneity between groups.

Example: Identifying Professors for Machine Learning

• Criteria: Professors should have knowledge in both statistics and computer science.
• Data: Research publications of professors in various subjects.
• Clustering Analysis: Groups papers into three buckets: Statistics, Computer Science, Machine Learning (combination).
• Outcome: Identifies professors capable of handling machine learning based on research areas.

Types of Clustering Techniques

Partitioning Method

• Algorithms: K-means, K-medoids.
• Cluster Center: Based on distance from data points to cluster center.
• Shape: Finds mutually exclusive, spherical or near-spherical clusters.
• Dataset Size: Effective for small to medium-sized datasets.

Hierarchical Method

• Structure: Creates hierarchical or tree-like structures via decomposition or merger.
• Distance Measurement: Based on nearest or farthest data points in neighboring clusters.
• Correction: Erroneous merges/splits can't be corrected at subsequent levels.

Density-Based Method

• Shape: Identifies arbitrarily shaped clusters based on object density.
• Guiding Principle: Dense regions of objects become clusters, filters out outliers.

Summary

• Covered Topics: Definition of clustering, applications, clustering as a machine learning task, types of clustering techniques.
• Next Class: Detailed discussion on partitioning method.