ELUC: evaluate predictor uncertainty

[ofrancon]

Train a model that evaluates point predictions from a predictor model.

For a particular prediction, if the model has been trained on plenty of similar features we can be more confident in the prediction than if the model has never seen this kind of features (out of distribution sample).

See for example Quantifying Point-Prediction Uncertainty in Neural Networks via Residual Estimation with an I/O Kernel for one methodology that could be used to train an uncertainty model. The associated code is available here

Other kind of models could be used too.