Hybrid Ayurveda using Machine Learning for Disease Prediction System using Dosha-Guided Feature Weighting

Abstract

Ayurveda is focused on a personal diagnosis along with the balance of doshas, but it is not an easy task to convert the qualitative concepts of Ayurveda into the decision-support systems. In this study, an Ayurveda-based prediction system of disease is presented with the help of the Machine Learning techniques such as Linear Regression (LR), Decision Tree (DT) and Dosha-Guided Feature Weighting (DGFW). The proposed model combines clinical information, lifestyle parameters, the symptom profile and the answers of Ayurvedic questionnaires in order to create a structured dataset to classify the disease. Throughout the preprocessing, Ayurvedic specific elements like Prakriti, Agni, Nadi and signs of imbalance in dosha go through normalization and weighting depending on the contribution to a particular disease. A DT model is used to learn the non-linear decision rule, which depicts Ayurvedic diagnostic logic, whereas Linear Regression is used to capture the linear relationships between weighted features. Theoretical testing on a multi-class data set shows that DT model performs better than LR, reaching an accuracy of 92.6 and LR reaches an accuracy of 88.4. Dosha-guided feature weighting to a large extent enhances classification performance, especially on non-linear models. The analysis of a confusion matrix and performance measures prove a decrease in misclassification and equal precision, recall and F1-scores by disease classes. The findings confirm the usefulness of integrating Ayurvedic domain knowledge with machine learning to provide a clear, understandable and effective decision-support system to predict diseases early and provide personalized healthcare.