Method and related apparatus for estimating a mental state, in particular a workload, of a subject, based on a stepwise linear discriminant analysis with automatic stop, that is capable to automatically optimise the process of selection of the features by using the brain activity of the subject the mental state of whom is to be estimated (e.g. the mental workload), in order to make the performance of classification of the considered mental state stable over time (in particular, over different days, optionally over at least one week). The method takes into consideration the EEG theta and alpha rhythms and introduces an automatic stop in the standard SWLDA method that automatically interrupts the process of selection of the features as soon as an optimal number of features have been added to the model.
