AI-Based Dosha Classification Models

Abstract

Dosha classification, the assignment of a patient to one of seven prakriti categories (Vata, Pitta, Kapha and four combination types), sits at the methodological foundation of personalised Ayurvedic practice. Classical prakriti assessment is carried out by trained practitioners through structured observation and interview, and has known inter-rater variability in the 60-78% range even among experienced practitioners. Self-administered prakriti questionnaires have been developed to improve consistency but have their own validation limitations. AI-based classification models have emerged as a methodologically distinct alternative that could improve consistency and scalability while preserving the clinical information value of prakriti classification. This study reviews 48 primary studies published between 2010 and 2021 evaluating AI-based dosha classification approaches, grouped under five model approach frames: D1 (questionnaire-based classical machine learning), D2 (biometric single-modality deep learning), D3 (multi-modal fusion approaches), D4 (ensemble and semi-supervised methods), and D5 (explainable AI with classical reasoning integration). Each frame was scored with the Dosha Classification Model Index (DCMI). Multi-modal fusion (D3) produced the strongest evidence at DCMI 0.832, followed by explainable AI (D5) at 0.814 and ensemble methods (D4) at 0.796.