This paper presents an improved just-in-time learning based random mapping partial least squares for soft sensor development.

How are the probabilities in Eq(5) determined? Some details should be given.

Eq(6) works out the mutual information between the input variables and output variables based on all the d output variables. In soft sensor development, instead of a multi-input multi-output model, several multi-input single output models can be developed. In such cases, Eq(6) may not be very relevant.

How many local relevant samples are selected and the impact on just-in-time learning should be discussed.