Model Accuracy (Analysis)

Source: (inline in performance/page.tsx)

---

What You See

Shows how accurate the PIRX prediction model is globally: Mean Absolute Error (in seconds), a badge showing whether it meets the accuracy benchmark, bias, 95% confidence limits, and sample size. If you have race results, a table compares your actual race times with what PIRX predicted.

How the Numbers Work

Mean Absolute Error (MAE) is the average difference between what PIRX predicted and what actually happened in real races. A lower number means better accuracy. Bias shows whether PIRX tends to predict too fast (negative) or too slow (positive). The 95% limits show the range where 95% of predictions fall.

Step by Step

Models & algorithms used: Bland-Altman

A monthly Celery task runs a Bland-Altman analysis comparing predictions to verified race results across all users. Mean Absolute Error (MAE) = mean of the absolute differences between predicted and actual times. Bias = mean of (predicted minus actual) -- positive means PIRX tends to predict slower than reality. 95 percent limits of agreement = bias plus or minus 1.96 times the standard deviation. Sample size = count of verified race results used. If the absolute bias exceeds 10 seconds, it is logged to the model_metrics table and triggers a bias-correction flag for the projection engine. If you have personal race results, the page shows a table with each race's date, actual time, predicted time, and the difference.