Explainable Machine Learning Using Sensor-Derived Biomechanical Features to Classify Elevated VALR-Related Loading...

Area of Science:

• Biomechanics
• Pediatric sports science
• Running mechanics

Background:

• Midsole hardness impacts lower-limb loading in pediatric forefoot strike (FFS) running.
• The biomechanical basis for identifying elevated Vertical Average Loading Rate (VALR) is not well understood.

Purpose of the Study:

• To investigate the relationship between midsole hardness and lower-limb impact loading during FFS running in children.
• To identify biomechanical features that characterize elevated VALR-related loading.

Main Methods:

• Fourteen boys ran in shoes with varying midsole hardness (37-52 Shore C).
• Lower-limb kinematics and sEMG data were collected; VALR and 28 biomechanical features were extracted.
• XGBoost machine learning models classified elevated VALR, with SHAP analysis identifying key features.

Main Results:

• VALR showed an increasing trend with midsole hardness, but no distinct threshold was found.
• XGBoost achieved 75.93% accuracy, with an AUC of 0.738.
• Distal and non-sagittal kinematic features were most influential in classifying elevated loading.

Conclusions:

• Elevated VALR-related loading in children's FFS running is characterized by a complex pattern of biomechanical features.
• A multi-feature model is more effective than a fixed midsole hardness threshold for identifying increased loading.