This paper presents and validates new formulas for estimating human body fat content, developed by Size Stream, using data from their SS20 3D body scanner and manual measurements.
The formulas offer a high level of accuracy and practicality, balancing statistical robustness with ease of application.
Key takeaway
Size Stream developed its SS20 body fat formulas using the four-compartment model across 1,790+ scans, finding that optimal predictors shift fundamentally between lean and high-adiposity individuals, an insight only discoverable through machine learning on a large, diverse dataset. The resulting formulas achieve R² > 0.9 for fat mass prediction with a mean difference of just 0.13 lbs, and held up under independent cross-validation at Pennington Biomedical Research Center and UC San Francisco/University of Hawaii. Critically, total body surface area, a variable only a 3D scanner can accurately capture, emerged as a key predictor for higher-adiposity individuals, underscoring why the SS20 provides a qualitatively different class of body composition data than any tape measure or photograph-based method.