Our paper “Personality cannot be predicted from the power of resting state EEG” was just accepted to Frontiers in Human Neuroscience. The hard-working researchers from the Institute of Psychology had collected a EEG data from almost 300 subjects and on top of that measured the personality of each subject. We applied machine learning and tried to read out from the EEG the personality characteristics of the participants. Think about it. It would have been relatively sweet if we could infer from the brain recordings that you are an extravert, who is open to experience and on the same time relatively neurotic! Anyway, we did not succeed. After our work we are in fact even quite sure that it is not possible to decode personality from EEG, that’s why we used such a strong statement “cannot be predicted” in the title of our paper. Let’s look closer at the reasons for this claim:
1. We had a very large sample size: in total the EEG data of 289 subjects were recorded and used. It is of course possible that with the data of 2000 subjects some information about the personality of the subjects could be decoded. Let’s wait and see.
2. The efficient nested cross-validation pipeline we applied used almost all of the data for hyper-parameter selection, for fitting the model and for estimating accuracy and statistical significance of the predictions. In other words, we made use of the data we had to take our best shot at decoding some information. Yet we found nothing.
3. We used lots of different methods to find the result: different machine learning algorithms, preprocessing tricks, normalization methods, dimensionality reductions etc. In total, 648 combinations of hyper-parameters were explored for each personality trait. In short, we did a lot of tricks to be able to infer personality from resting state EEG but received a negative result.
4. We DO believe that it is practically impossible to decode personality from resting state EEG. We believe so until someone proves otherwise.