Machine learning aided jump height estimate democratization through smartphone measures.

Area of Science:

• Biomechanics
• Sports Science
• Wearable Technology

Background:

• Countermovement jump (CMJ) height is a key performance metric.
• Traditional CMJ assessment relies on force platforms or inertial sensors.
• Smartphones offer a potential low-cost alternative for CMJ height estimation.

Purpose of the Study:

• To develop and validate a smartphone-based method for estimating CMJ height.
• To compare the accuracy of smartphone-derived CMJ height with gold-standard measurements.
• To identify key biomechanical features for accurate CMJ height prediction.

Main Methods:

• 43 participants performed CMJs holding a smartphone.
• Smartphone inertial sensor data was collected alongside force platform data.
• A multi-layer perceptron model was trained using Lasso regularization for feature selection.
• Hyperparameter optimization was performed using grid search with cross-validation.

Main Results:

• The developed smartphone method achieved high accuracy (4 cm error) and precision (4 cm error).
• This significantly outperformed raw smartphone data estimates (18 cm error, 16 cm precision).
• Peak acceleration and braking phase duration were identified as crucial predictive features.

Conclusions:

• Smartphone-based CMJ height estimation is feasible and accurate.
• Machine learning enhances the predictive power of smartphone inertial data.
• This approach democratizes access to reliable jump performance analysis.