Unsupervised Learning

• Unsupervised learning uses unlabeled data for training.
• This is particularly useful in problems where it is difficult to obtain labeled data.
• Unsupervised learning algorithms are useful for discovering patterns in data.
• The goal of unsupervised learning is use a feature vector as input and either outputs a feature vector or a label/value.

Use Cases

• Clustering algorithm takes a feature vector as input and outputs a label.
• Dimensionality reduction algorithms take a feature vector as input and output a feature vector with less features.
• Anomaly detection algorithms take a feature vector as input and output is a real number indicating the degree of anomaly.

Examples

• Clustering:
  • k-Means
  • Hierarchical Clustering
  • DBScan
• Anomaly Detection:
  • Isolation Forest
  • One-class SVM
• Dimensionality Reduction:
  • Principal Component Analysis
  • t-Distributed Stochastic Neighbor Embedding
  • Kernel PCA
  • Locally Linear Embedding
• Association Rule Learning:
  • Apriori
  • Eclat