Mindful learning: meditative state classifier using random forest

PALLAVICINI, CARLA; TAGLIAZUCCHI, ENZO; MARTINEZ VIVOT, ROCIO

Huerta Grande

Congreso; XXXIIIº Annual Congress of the Argentinean Society for Neuroscience Research; 2018

Sociedad Argentina de Neurociencias

The use of EEG signals provides a promising analysis to assess the meditative state in an objective fashion. The changes in brain activity during the meditative state are well documented, however models that build classifiers in a way that enables them to distinguish the meditative state are scare. This analysis was performed in 3 groups of meditators: Himalayan Yoga (HT), Vipassana (VIP) and Isha Shonnya Yoga (SNY). Briefly, the random forest method is based on a decision tree concept where each node represents one characteristic and classes are divided aiming for the maximum purity of the data until 100% purity is reached, hence that group of data becomes classified. To analize the similarities between the different meditative states in a multivariated way, random forest classifiers were generated in order to differentiate the cerebral states of each from the power potential of each spectral band. Throughout all the spectrum studied the trained classifiers with HT data were able to distinguish correctly the SNY data and for all the bands except one the optimized classifiers to distinguish VIP generalized the data of SNY. We observe a similarity between the states along all the bands, being δ, low γ and high γ the ones that present a higher classification inter-state. These results suggest that there is common ground in the cerebral states reached through different meditative traditions, represented by the EEG data, which manifest throughout the studied frequency spectrum.