Supervised learning learns a relationship between labeled inputs and known outcomes, then applies that relationship to predict outcomes for new, unseen data. section 4.0
Choose the supervised-learning method according to the target and the relationship being modeled:
- Use linear regression when the response is numeric and the relationship between predictors and outcome is approximately linear. section 4.0 section 4.1
- Use classification methods when the outcome is categorical, such as a label or group membership; do not use linear regression for this task. section 4.1
- Evaluate predictions with held-out data; decision trees and random forests use test sets, accuracy, and confusion matrices to assess generalization. section 6.5
Prediction and decision support
Applications include predicting sales from advertising expenditure, home prices from property characteristics, student performance from academic measures, and product demand from historical sales. section 4.0
