Machine Learning Basics: Supervised vs Unsupervised

• Supervised Learning (labeled data)
  • Regression: predict a number
    • Linear regression, e.g. house prices
  • Classification: predict a category
    • Logistic regression, decision trees, random forests
  • Requires labeled training examples
• Unsupervised Learning (no labels)
  • Clustering: k-means, DBSCAN
  • Dimensionality reduction: PCA
  • Finds hidden structure in raw data
• Training Essentials
  • Split data: train / validation / test
  • Overfitting: memorizes training data, fails on new data
  • Underfitting: model too simple for the pattern
  • Cross-validation gives honest estimates
• Evaluation Metrics
  • Accuracy misleads on imbalanced classes
  • Precision vs recall: a real trade-off
  • RMSE is the classic regression error measure
• Classic Starter Datasets
  • Titanic: binary classification
  • Iris: multi-class classification
  • Boston-style housing data: regression