A predicted value is the outcome a model estimates for particular input values. Depending on the model, it may be a probability, such as the chance of a meltdown, or a numerical ŷ-value produced by a regression line. section 8.2 section 7.9
Predicted values play two closely related roles:
- Logistic regression predicts a probability from input data. A threshold, commonly 0.5, can convert that probability into a yes-or-no classification. section 8.2
- Linear regression predicts a numerical value from a line. Comparing each predicted ŷ-value with its actual y-value produces a residual. section 7.9
- Squaring and adding residuals produces the sum of squares, a loss measure where smaller values indicate that predictions lie closer to the data overall. section 7.9
A nap-time prediction
Suppose a toddler has gone 4.78 hours without a nap. A fitted logistic curve estimates a 97% probability of an imminent meltdown. That probability is a predicted value—not a guarantee—so a chosen threshold determines whether the model labels the outcome “meltdown” or “no meltdown.” section 8.2
