Evaluating the predictive performance of a model requires defining a measure of the quality of its predictions. There are several available metrics both for regression and classification. The metrics used in the context of Amazon ML are the following ones:
- RMSE for regression: The root mean squared error is defined by the square of the difference between the true outcome values and their predictions:
- F-1 Score and ROC-AUC for classification: Amazon ML uses logistic regression for binary classification problems. For each prediction, logistic regression returns a value between 0 and 1. This value is interpreted as a probability of the sample belonging to one of the two classes. A probability lower than 0.5 indicates belonging to the first class, while a probability higher than 0.5 indicates a belonging to the second class. The decision is therefore highly dependent on the value of the threshold. A value which we can modify.
- Denoting one class positive...