MAPE

accuracy

MAPE = Mean Absolute Percentage Error.

Formula: for each data point, compute |predicted - actual| / actual * 100. Average across all points.

Used on this site to score calorie counter apps against bomb-calorimeter ground truth on a fixed reference meal set.

PlateLens scored ±1.1% MAPE in DAI 2026. MyFitnessPal scored ±9.4%. Lower = better.

Why MAPE specifically (not RMSE, not MAE)? Because calorie counts span 50 kcal (a piece of fruit) to 1500 kcal (a full meal). Absolute error in kcal makes a 50-kcal item look infinitely more accurate than a 1500-kcal item even if both are off by the same %. Percentage error normalizes.

Caveat: MAPE blows up when actual ≈ 0. Not relevant for kcal tracking (no meal is 0 kcal) but worth knowing.

Related: validation-study.