Predicting Ayurveda-Based Constituent Balancing in Human Body Using Machine Learning Methods

Human Body constitution (
prakriti
) defines what is in harmony with human nature and what will cause to move out of balance and experience illness.
Tridosha
defines the three basic energies or principles that determine the function of our body on the physical and emotional levels. The three energies are known as
VATT
,
PITT
and
KAPH
. Each individual has a unique balance of all three of these energies. Some people will be predominant in one, while others will be a mixture of two or more.
Ayurveda-dosha
studies have been used for a long time, but the quantitative reliability measurement of these diagnostic methods still lags behind. A careful and appropriate analysis leads to an effective treatment. To collect a meaningful data set, a questionnaire with 28 different characteristics is validated by Ayurveda experts. Authors calculate Cronbach alpha of
VATT-Dosha
,
PITT-Dosha
and
KAPH-Dosha
as 0.94, 0.98 and 0.98, respectively to check the reliability of the questionnaire. Authors analyzed questionnaires of 807 healthy persons aged 20-60 years and found 62.1% men and 37.9% women. The class imbalance problem is resolved with oversampling and the equally distributed data set of randomly selected 405 persons is used for the actual experiment. Using computer algorithms, we randomly divide the data set (8:2) into a training set of 324 persons and a test data set of 81 persons. Model is trained using traditional machine learning techniques for classification analysis as Artificial Neural Network (ANN), K-Nearest Neighbour (KNN), Support Vector Machine (SVM), Naive Bayes (NB) and Decision Tree (DT). System is also implemented using ensemble of several machine learning methods for constitution recognition. Evaluation measures of classification such as root mean square error (RMSE), precision, recall, F-score, and accuracy is calculated and analyzed. On analyzing the results authors find that the data is best trained and tested with CatBoost, which is tuned with hyper parameters and achieves 0.96 precision, 0.95 recall, 0.95 F-score and 0.95 accuracy rate. The experimental result shows that the proposed model based on ensemble learning methods clearly surpasses conventional methods. The results conclude that advances in boosting algorithms could give machine learning a leading future.