Unsupervised Learning

  • After supervised learning, the most common form of machine learning is unsupervised learning. In unsupervised learning, we are given data without any output labels \(y\).
  • Data comes with inputs \(x\) but no outputs \(y\) and the algorithm has to find structure in this data.
  • Let’s take our earlier breast cancer prediction problem for an example.

  • We’re not asked here to predict whether the tumor is malignant or benign because we are not given any labels of which tumor is which.
  • Instead, our job is to find some pattern, or some data, or just something interesting within this unlabeled dataset.
  • The reason this is called unsupervised learning is that we are not asking the algorithm to give us a “right answer”.
  • In this example, our unsupervised algorithm might decide there are two clusters, with one group here and one there.

  • This is a specific type of unsupervised learning algorithm called clustering algorithm because it places the unlabelled data into different clusters.

  • Clustering algorithm

  • Clustering groups similar data points together.
  • Clustering has many use cases:
    • It is used in Google News! Google News looks at 100’s of stories every day and clusters them together.

    • It is used in DNA microarray clustering. The red here might represent a gene that affects eye color, or the green here is a gene that affects how tall someone is.
      • You can run a clustering algorithm to group different types of individuals together based on categories the algorithm has automatically decided.
    • It is used in grouping customers in different market segments to better understand a company’s consumer base. This could help in improving marketing strategies for each group.

Anomaly detection algorithm

  • Find unusual data points. This could be used to detect fraud in bank transactions.

Dimensionality reduction

  • Compress data using fewer numbers.