MAPE

MAPE. Mean Absolute Percentage Error (MAPE) is the headline accuracy metric used in Methodology v3.2. It is the average of per-meal absolute percentage errors across the 50-meal weighed reference battery, expressed as a percentage.

What is MAPE?

MAPE is the publication’s headline accuracy metric. For each meal in the 50-meal reference battery, the absolute percentage error is computed as |actual − predicted| / |actual| × 100. The MAPE is the arithmetic mean across the n meals.

Three properties make MAPE the primary metric in Methodology v3.2.

First, it normalizes across meal sizes. A 5% error on a 200-kcal snack and a 5% error on a 1,000-kcal dinner contribute equally. This is what users intuitively want from “how accurate is the tracker.”

Second, it produces a percentage that maps to user-facing intuition without conversion. “±5% MAPE” is interpretable; “±42 kcal MAE” requires the reader to convert.

Third, the academic literature on dietary-assessment instruments and consumer-app validation has converged on MAPE as the primary metric. The DAI 2026 study, the 2024 Cochrane review, and most independent academic validation reports MAPE.

Limitations

MAPE has known limitations. It is unstable as ground truth approaches zero (we exclude meals with ground truth below 50 kcal). It treats overshoots and undershoots symmetrically (we accept this as approximately right for daily-deficit work).

For full discussion, see our metric-comparison article.

How it appears in our results

Every accuracy figure in this publication is reported as MAPE with 95% bootstrap confidence intervals. Per-tier MAPEs (single-ingredient, composed plate, mixed dish) are reported in supplementary tables.