EEG study achieves 91% accuracy in classifying programmer skill levels

A new study published on arXiv (arXiv:2606.30879) demonstrates that electroencephalography (EEG) combined with machine learning can accurately classify a programmer's skill level. Led by Maurice Rekrut and collaborators, the team analyzed an existing EEG dataset recorded during code comprehension tasks from 37 programmers with experience ranging from 1 to 30 years (mean 8.1 ± 6.3 years). Using Random Forest classifiers, they examined both binary classification (experts vs. novices) and multi-class classification (experts, intermediates, novices). The binary classification achieved an impressive 91.83% average accuracy in stratified 10-fold cross-validation, while the multi-class setup reached 78.15%. Even with leave-one-subject-out validation—a stricter test that generalizes across individuals—binary accuracy remained high at 85%, with multi-class at 58.80%.

The researchers identified specific neural signatures correlated with programming expertise. EEG entropy (a measure of signal complexity) showed the strongest correlation with skill level. Notably, experts' brains exhibited highly localized centro-frontal activation, whereas novices showed more distributed frontal activation networks—suggesting that experienced programmers process code more efficiently and with less cognitive scatter. Individual frequency bands outperformed full-spectrum analyses, and both program comprehension and resting-state data yielded strong classification results. This research highlights the potential of neural data as an objective complement to traditional methods like coding tests or interviews, offering a new dimension for hiring, team composition, and performance evaluation in software development.