Logistic Regression

Extension of Linear Regression. Models probabilities for classification problems. Negative = 0, Positive = 1

Use Cases

  • Binary Classification
  • Customer Churn Prediction … whether a customer stays or leaves
  • Medical Diagnosis … predicting presences/absence of a disease

Advantages

  • Can include probabilities
  • Can be extended for multi-class classification

Disadvantages

  • Worse predictive performances than other models
  • Difficult interpretation
  • Can’t be trained if there is a feature perfectly separating the two classes