Motivating Question

GOALS

Suppose we have a set of feature variables \((x_1,x_2,...,x_k)\) but NO outcome variable \(y\).

Instead of our goal being to predict/classify/explain \(y\), we might simply want to…

1. Examine the structure of our data.
2. Utilize this examination as a jumping off point for further analysis.

UNSUPERVISED METHODS

• Cluster analysis
  • Focus: Structure among the rows, i.e. individual cases or data points.
  • Goal: Identify and examine clusters or distinct groups of cases with respect to their features x.
    
  • Methods: hierarchical clustering & K-means clustering
• Dimension reduction
  • Focus: Structure among the columns, i.e. features x.
  • Goal: Combine groups of correlated features x into a smaller set of uncorrelated features which preserve the majority of information in the data. (We’ll discuss the motivation later!)
  • Methods: Principal components

CLUSTERING vs DIMENSION REDUCTION EXAMPLE

CLUSTERING EXAMPLES

“Love Actually” analysis

Machine learning about each other

Identify genetic similarities among a group of patients

Cartograph.info