Chapter summary¶
This chapter presents a review of regression models with predictive uncertainty which can be used to describe the relationship between variables in engineering models, in addition to describing how the models can be created in practice from data. We reviewed probabilistic models and non-probabilistic models. Probabilistic regression models use probability distributions to express information about the variability and uncertainty in the modelled output; they are currently the most widely used regression models. Non-probabilistic models are useful in cases where only limited or imprecise data may be available, and prior knowledge of regression model parameters may be difficult to obtain. A particular advantage of Convex IPMs are the a priori bounds on the model bound violation, which can be used to validate the model at training time without test data.