Classifying creativity: Applying machine learning techniques to divergent thinking EEG data

Abstract

Prior research has shown that greater EEG alpha power (8–13 ​Hz) is characteristic of more creative individuals, and more creative task conditions. The present study investigated the potential for machine learning to classify more and less creative brain states. Participants completed an Alternate Uses Task, in which they thought of Normal or Uncommon (more creative) uses for everyday objects (e.g., brick). We hypothesized that alpha power would be greater for Uncommon (vs. Common) uses, and that a machine learning (ML) approach would enable the reliable classification data from the two conditions. Further, we expected that ML would be successful at classifying more (vs. less) creative individuals. As expected, alpha power was significantly greater for the Uncommon than for the Normal condition. Using spectrally weighted common spatial patterns to extract EEG features, and quadratic discriminant analysis, we found that classification accuracy for the two conditions varied widely among individuals, with a mean of 63.9%. For more vs. less creative individuals, 82.3% classification accuracy was attained. These findings indicate the potential for broader adoption of machine learning in creativity research.

Keywords

• Alpha
• Classification
• EEG
• Creativity
• AUT