Predicting Running Performance and Recovery Using Wearable Data: An N-of-1 Study

This longitudinal N-of-1 case study investigated the associations between daily physiological recovery, lifestylebehaviors, and running training performance. A single male runner (weight: 80.0 kg) was tracked continuously for 196days using a WHOOP strap (collecting recovery score, heart rate variability [HRV], resting heart rate [RHR], sleep logs,and a lifestyle journal) and Strava (collecting running metrics), yielding 37 running training days. Ordinary Least Squares(OLS) robust regression (HC3 covariance estimator) and cross-validated machine learning models (Random Forest,XGBoost) were evaluated. For running performance (Grade-Adjusted Pace [GAP]), morning autonomic metrics showedno significant linear associations (p > 0.05). For next-day recovery, cardiovascular strain was significantly associatedwith higher next-day recovery (β = 1.360% per unit, p = 0.0010), whereas run distance was negatively associated (β = -1.595% per km, p = 0.0094). Lifestyle factors showed positive recovery trends, such as taking a cold shower (β =+16.95%, p = 0.126) and consuming carbohydrates (β = +17.41%, p = 0.296). OLS outperformed machine learningmodels, which overfit this small-sample time series. Daily training pace is associated with training volume rather thandaily recovery, whereas next-day recovery is significantly associated with training loads.