Mean Absolute Percentage Error (MAPE)
MAPE is the standard metric for measuring how far an app's predicted calorie value diverges from a known reference value, expressed as a percentage.
Mean Absolute Percentage Error (MAPE) is the average of the absolute differences between predicted and reference values, expressed as a percentage of the reference. For calorie tracking app accuracy testing, the formula is:
MAPE = mean(|predicted_kcal − reference_kcal| / reference_kcal) × 100
A 5% MAPE means an app’s calorie estimates are on average within ±5% of the weighed reference. We use MAPE as the headline accuracy metric on every review and ranking because it is interpretable, symmetric in error direction, and comparable across apps that use different food databases.
Why MAPE and not raw error
Raw error (kcal off) penalizes large meals more than small ones. Percentage error normalizes for portion size. MAPE is the metric used in academic dietary-assessment validation literature (Schoeller 1995; Subar 2015), so using it here makes our numbers comparable to peer-reviewed work.
Known limits
MAPE treats over- and under-estimation as equally bad. An app that systematically reports 10% high on every meal has the same MAPE as an app that randomly varies ±10%, even though the first is correctable with a simple bias adjustment and the second is not. We report MAPE alongside the directional bias when the bias is non-trivial.
MAPE is also undefined or unstable when reference values approach zero — a tracker estimating 5 kcal on a true 1 kcal reference produces a 400% MAPE that is misleading on absolute terms. For calorie tracking this rarely matters; portion sizes large enough to track are large enough to compute MAPE stably.
How we use MAPE in our reviews
Our scoring rubric uses MAPE as the anchor for the Accuracy & Database criterion (25% of the 100-point total). The formula:
accuracy_points = clamp(100 − MAPE × 4, 0, 100)
So 5% MAPE → 80 points, 15% MAPE → 40 points, 25%+ → 0 points. The slope was chosen so an app at the boundary of clinical usefulness (~5% MAPE per Schoeller 1995) gets a strong but not perfect score.