Loading subject…
Model comparison
The evidence that supports a model selection, gathered by training each shortlisted candidate on the same data split and scoring every one of them on the same metric.
Example
Two churn models are trained on the same split and scored on recall, giving 0.62 and 0.78, so the margin between them is 0.16.
Model selection
The decision about which machine learning algorithm to use on a problem, made by matching the algorithm's assumptions to the required output, the characteristics of the data and the resources available.
Example
A hospital with 300 labelled scans selects a decision tree rather than a deep network, because a network of that size would overfit so little data.