Machine learning-based equations for improved body composition estimation in Indian adults

This includes: (S1) a list of data quality rules applied for inclusion in the study; (S2) performance metrics (compared with DXA-based measurement) for different prediction algorithms using all predictors in training data; (S3) Performance metrics (compared with DXA-based measurement) for different prediction algorithms using all predictors in test data, overall and stratified by age (<40 years test n = 122 female and 185 male, 40 + years test n = 167 female and 133 male); (S4) Performance metrics (compared with DXA-based measurement) for different prediction algorithms using all predictors in test data, based on 25 datasets where the DXA-based outcomes were randomly permuted (to provide a null or baseline scenario to compare against performance on the real data); (S5) Performance (Mean Absolute Error) of the LASSO with alternate sets of predictors; (S6) Coefficients for each outcome in full model; (S7) Information about model performance for other outcomes (trunk fat mass (kg), trunk lean mass (kg), L1-L4 fat mass (kg), L1-L4 lean mass (kg), appendicular fat mass (kg), and appendicular fat mass percentage (%)); and (S8) Minimum, 1st percentile, 99th percentile, and maximum of each predictor in the training data.