Abstract: Highlights•We construct a generalization of RBF kernel to incomplete data.•It is the expected value of classical RBF.•We do not perform any imputations but model missing values by probability distributions.•It is easy to implement and can be used together with any kernel method.
Loading