Impact of Yoga Practices on Cognitive Functions through a Machine Learning Approach: A Protocol for Evidence Mapping and Predictive Modelling

Abstract

Yoga is increasingly used as a research-supported, complementary approach for improving cognitive, psychological, neural and physiological health, and long-term practice is reported to produce qualitatively different outcomes from short-term practice. However, the existing literature is characterised by substantial heterogeneity in intervention type, duration, dosage, study population and outcome assessment, which limits comparability across studies. This paper sets out a systematic evidence-mapping protocol for cataloguing the long-term neural, cognitive, psychological and physiological effects of yoga, together with pre-specified eligibility criteria, intervention and outcome definitions, and an evidence-synthesis procedure. Building on this protocol, the paper further proposes a machine-learning (ML) framework intended to predict cognitive improvement from intervention type and duration, and reports an illustrative proof-of-concept classification analysis — using a labelled yoga-posture dataset and a Decision-Tree-based model benchmarked against Logistic Regression, Extreme Gradient Boosting and Random Forest — together with the data requirements, preprocessing and validation strategy that would apply to the full predictive model. The proposed Decision Tree model achieved the highest performance among the models compared (accuracy 0.98, precision 0.9771, recall 0.91, F1-score 0.89) on the illustrative dataset. The paper discusses the research-design and execution challenges inherent in synthesising heterogeneous yoga protocols and in training reliable predictive models on such data, including publication bias, between-study heterogeneity, sample-size requirements for ML, overfitting risk, external validation, model interpretability, and the distinction between association-based prediction and causal inference. The integration of evidence mapping with machine learning is presented as a framework for identifying patterns between yoga-intervention characteristics and cognitive outcomes that may, in future, support personalised mind–body interventions.