What criteria are used for models estimation and ranking in PNN Discovery Client?
There are two criteria used for models estimation. First, RSS – quadratic deviation calculated over learning subset. Second, RR – quadratic deviation calculated over test subset. There are two additional external criteria which can be used for model estimation, but aren’t used for models ranking: correlation coefficients R2 and cross-validated R2 (well-known as Sibyl). For more information, see tutorial.
