The Application of Machine-Learning Approach for the Classification of People According to Their Participation in Meditation based on Neurophysiological Data

Motor control is a human ability to manage own goal-directed motions. This ability can be estimated by means of so-called stop-signal paradigm (SSP). SSP is an experimental approach consisting of two behavioral tasks - activation or inhibition of motions. SSP is used for diagnosing a wide range of neuropsychiatric pathologies, including depressive and anxiety disorders. Combined with the analysis of evoked brain potentials (ERP), processing of the behavioral SSP results allows to reveal the neurophysiological causes of normal motor control and its deviations. Meditation is a psychological practice aimed at reducing stress and anxiety levels. The result of longterm meditation is the increase of individuals’ resilience to affective disorders. In this study, we applied a machine-learning approach to develop a methodology of classification of people according to their participation in meditation. The amplitudes of two ERP peaks (premotor and postmotor) were used as input data. We developed four convolutional network models that were trained and tested on approximately 100 healthy participants (half of whom participated in meditation). Then, all models were checked on an additional sample of 25 participants. We selected parameters for convolutional networks that allowed us to achieve $\mathbf{8 2 \%}$ classification accuracy and model stability against overfitting. The proposed approach allows to classify individuals based on their stress resilience through ERP data processing.