A discriminative model predicts outcomes by learning distinctions between classes or conditional outcomes. It creates a boundary separating categories and estimates the probability of an output given an input. section 10.3
Discriminative models commonly support supervised learning tasks such as:
- Classifying whether a customer will churn or whether a credit-card transaction is fraudulent. section 10.3
- Estimating numeric outcomes, such as next month’s store sales or a call center’s expected call volume. section 10.3
- Applying logistic regression, decision trees, random forests, support vector machines, or deep learning networks to prediction and classification. section 10.3
Classification versus generation
Given labeled transaction data, a discriminative model learns to distinguish fraudulent from genuine transactions. A generative model instead learns the broader data distribution and can create new data points resembling its training data. section 10.3
