On intelligent Prakriti assessment in Ayurveda: a comparative study.

This article focuses on applying and comparing eight supervised machine learning (ML) classifiers to predict Prakriti, the Ayurvedic constitution types, using phenotypic data from two genetically homogeneous cohorts in northern and western India. Prakriti classification, central to Ayurveda's holistic health management, categorizes individuals into seven types based on physical and physiological traits, with three extreme types—Vata, Pitta, and Kapha—being the primary focus here. The study evaluates the classifiers' performance across multiple metrics—including accuracy, precision, F1-score, AUROC, Matthews correlation coefficient, and Hamming loss—using cross-validation and cross-cohort testing, finding that multinomial naïve Bayes classifier (MNBC), ν-support vector classifier (ν-SVC), and extra trees classifier (ETC) perform best for the northern cohort, western cohort, and cross-cohort predictions, respectively. The research highlights that reduced feature sets selected via recursive feature elimination can effectively train ML models for Prakriti prediction, suggesting potential for integrating artificial intelligence with traditional Ayurvedic assessment to enhance diagnostic consistency and personalized treatment.