Predicting knee pain severity in professional distance runners using gaussian process regression

Background: Knee pain is common in runners and may precede injury; clinical labels are sparse, but repeated symptom monitoring is practical for training decisions. The purpose of this paper is to develop a probabilistic, interpretable model that predicts a continuous knee pain severity score in professional distance runners using Gaussian Process Regression (GPR), and to quantify the relative influence of routinely measurable training and anthropometric variables. Methods: Fifty professional male distance runners aged 20-30 years were recruited from professional athletics in Ankara, Turkey. Six predictors were used: resistance training frequency, weekly running distance, training surface, running technique (foot-strike category), history of prior knee problems, and body mass index (BMI). The outcome was a continuous knee pain severity score normalized to 0-1 from a self-reported numerical rating scale. A GPR model with a radial basis function kernel was trained using a train-test split and evaluated with mean squared error (MSE). Results: Prediction uncertainty was quantified using 95% credible intervals, and sensitivity analysis were performed to assess feature influence. The GPR model achieved low prediction error on the held-out test set (MSE= [0.02]) and stable performance under cross-validation (mean MSE= [0.028]). Conclusion: Sensitivity analysis identified resistance training frequency and BMI as the strongest predictors of knee pain severity, with smaller contributions from weekly running distance and prior knee injuries. Training surface and running technique had minimal effects. GPR effectively models self-reported pain with uncertainty quantification and is scalable for symptom-based risk assessment, though not intended for clinical diagnosis.