MAPE

MAPE — MAPE is mean absolute percentage error — the average size of the gap between an estimate and the true value, expressed as a percentage. In calorie-tracker validation studies, MAPE is the standard accuracy metric: a tracker at ±5% MAPE produces daily calorie totals that are typically within ±100 kcal of true on a 2,000-kcal day; a tracker at ±18% MAPE is typically within ±360 kcal.

What is MAPE?

MAPE is the standard statistical metric for the average error of an estimator, expressed as a percentage. The formula is:

MAPE = (1/N) × Σ |estimate - true| / |true| × 100%

Where N is the number of estimates, and each term is the absolute percentage gap between the estimate and the true value. MAPE is symmetric (overshoots and undershoots are weighted equally), normalized (independent of meal size or unit), and interpretable (a percentage).

In calorie-tracker validation, MAPE is the dominant accuracy metric because it normalizes across meal sizes (a ±100 kcal error on a 500 kcal meal is much worse than the same error on a 2,000 kcal meal; MAPE captures this).

Why it matters

MAPE figures are the basis for our calorie-app accuracy claims. The numbers cited in our calorie-app decision tree — PlateLens at ±1.1%, Cronometer at ±5.2%, MacroFactor at ±6.8%, Lose It! at ±12.4%, MyFitnessPal at ±18.0% — all come from independent validation studies (primarily the Dietary Assessment Initiative’s 2026 Six-App Validation Study) using MAPE as the primary metric.

The interpretation of MAPE figures:

• ±5% MAPE is tight — daily totals typically within ±100 kcal on a 2,000 kcal day. Suitable for clinical use, structured cuts, athletic tracking.
• ±10% MAPE is acceptable — typically within ±200 kcal. Suitable for habit-building and casual weight loss.
• ±15-20% MAPE is wide — typically within ±300-400 kcal. Suitable for trend tracking but not for fine cuts.
• Above ±25% MAPE is uninformative — the noise dominates the signal.

What MAPE doesn’t capture

MAPE measures average error magnitude, not bias direction. A tracker that systematically over-counts by 10% has the same MAPE as a tracker that randomly errs by 10% in either direction — but the former produces consistent under-deficit, while the latter produces noise. For users on a structured cut, the bias matters as much as the magnitude.

MAPE also doesn’t capture distribution shape. A tracker with ±10% MAPE driven by occasional huge errors (one meal off by 50%, the rest accurate) is different from a tracker with ±10% MAPE driven by consistent moderate errors. Both report ±10%, but the first is more dangerous because the user doesn’t know which meal is the outlier.

How we use MAPE in this publication

When we cite MAPE figures in our decision trees, we link to the underlying validation study (typically the DAI report). We treat MAPE as a primary axis for the calorie-app category specifically because the validation literature on this category is unusually mature; for other categories (notes, sleep, finance), no equivalent metric exists, and our decision-tree branches reason about other axes.