A novel symbolic regression-based approach for decoding the impact of meditation on cognitive enhancement using multimodal EEG signal analysis

Despite growing recognition of meditation's potential benefits, the underlying neural mechanisms and a comprehensive method to quantify its effectiveness remain elusive. Most studies focus on either electroencephalogram (EEG) metrics or cognitive performance separately, with limited attempts to combine both to assess the benefits of meditation. Few studies use machine learning to track meditation effectiveness over time through meditative state classification. Existing models often rely on limited EEG features and lack efficient feature selection, reducing classification accuracy and generalizability. To address these gaps, a two-week mantra meditation program was conducted for novices and experienced meditators. Participants underwent Brain-Based Intelligence Test (BBIT) assessments before (pre-week 1) and after (post-week 2) to measure changes in sustained attention, working memory, and cognitive flexibility. Weekly EEG data was collected throughout the program to explore underlying neural mechanisms. Symbolic regression identified formulas linking brainwave power to improvements in cognitive performance. Additionally, machine learning-based classification was applied to weekly EEG data using scalogram-derived features and complexity measures from visibility graphs. The simple, fast, and efficient (SFE) algorithm, utilizing exploration and exploitation phases, selected the best features for optimal performance. Classification accuracy improved across sessions, with XGBoost achieving 88.88% in session S1 and 100% in session S3 for experienced meditators, and 83.85% in session S1 to 94.73% in session S3 for novices. These findings demonstrate that short-term mantra meditation interventions induce distinguishable brain patterns and cognitive enhancements, providing a generalizable framework for evaluating meditation's impact on brain activity and cognitive function.