Research Masterclass in Radiotherapy Physics 2017

Validating Phenomenological models: a key point

- The ideal predictive model should:

Adequately fit the data (no under or overfit !) Giving correct predictions outside the original (training) data set on new (validation) data sets More external validations means more confidence in applying the model (…and negative results reduce it) Different degrees of generalization of the model based on the type of validation (range of doses, delivery technique, temporal, geographical,……) Heavy lacks of available predictive models with (satisfactory) external validation ! Need of much more efforts to accomplish accurate and well assessed validation studies

External validation