Prediction of spirometry parameters of adult Indian population using machine learning technology

Spirometry is one of the important non-invasive, sensitive, easy-to-perform, reproducible, and objective biomedical screening and diagnostic procedures in healthcare for the assessment of lung function. To date, there is no unified system, equation, or framework for the prediction of spirometry parameters for the Indian population. In this research article, a machine-learning-based system has been proposed and evaluated, and a web application developed for the prediction of Spirometry Parameters of the Adult Indian Population. The four most commonly used supervised machine-learning algorithms (Linear Regression, Gradient Boosting Regression, Deep Neural Multi-Layer Perceptron (MLP) Regression, and Support Vector Regression) for regression tasks have been evaluated for this purpose. Based on Mean absolute error, root mean squared error and adjusted R2 value, it has emerged that Gradient Boosting and Deep Neural MLP are the best-fit models to predict Forced Vital Capacity (FVC) and Forced Expiratory Volume in one second (FEV1) respectively for the Indian population. A web application has been designed using the Flask web framework to predict the FVC, FEV1, and corresponding Lower Limit Normality. This research work paves the foundation for ML-assisted spirometry for lung function assessment of the Indian population to extend the benefits of state-of-the-art technology in healthcare.