Overview

• Definition: Unsupervised machine learning involves training algorithms on data without labeled responses.
• Goal: Discover underlying patterns, structures, and anomalies in the data.
• *Applications:
  • Exploratory data analysis.
  • Used as a preprocessing step for supervised learning.

What is Clustering?

• Definition: Clustering is the task of dividing a set of data points into groups (clusters) such that points in the same group are more similar to each other than to those in other groups.
• Applications: Market segmentation, document clustering, image segmentation, etc.

K-Means Clustering

• Algorithm:

  1. Initialize cluster centroids randomly.
  2. Assign each data point to the nearest centroid.
  3. Recompute the centroids as the mean of all points in the cluster.
  4. Repeat steps 2-3 until convergence.

• Objective Function:

  where is the centroid of cluster .

What is Anomaly Detection?

• Definition: Identifying rare items, events, or observations that raise suspicions by differing significantly from the majority of the data.
• Applications: Fraud detection, network security, fault detection in systems.

New Methods

Machine Learning Methods

• Isolation Forest:
  • Randomly select a feature and split the data.
  • Anomalies are isolated quicker than normal points.

Clustering-Based Methods

• DBSCAN (Density-Based Spatial Clustering of Applications with Noise):
  • Points in low-density regions are considered anomalies.

What is PCA?

• Definition: PCA is a dimensionality reduction technique that transforms data into a new coordinate system.
• Goal: Reduce the number of dimensions while preserving as much variance as possible.

PCA Steps

1. Standardize the Data:
   • Ensure each feature has zero mean and unit variance.
2. Compute the Covariance Matrix:
   • Measure how features vary with respect to each other.
3. Compute Eigenvalues and Eigenvectors:
   • Determine the directions of maximum variance.
4. Project the Data:
   • Transform the data onto the new axes defined by the eigenvectors.

Applications of PCA

• Visualization:
  • Reducing data to 2 or 3 dimensions for plotting.
• Noise Reduction:
  • Removing components with low variance.
• Feature Extraction:
  • Combining original features into a smaller set of uncorrelated features.

Summary

• Clustering groups similar data points together.
• Anomaly Detection identifies data points that differ significantly from the majority.
• PCA reduces the dimensionality of data while preserving variance.

Conclusion

• Unsupervised machine learning techniques are powerful tools for discovering hidden patterns and structures in data.
• Clustering, anomaly detection, and PCA each have unique applications and methodologies.
• We can pair unsupervised learning with supervised learning to build more robust models.

Exercise

https://shorturl.at/J9hGc