Predicting Multi-planar Single-Leg Drop Landing Kinetic and Dynamic Stability Outcomes from Various Wearable Inertial Measurement Unit Configurations: A Machine Learning Approach with Implications for Field-Based Injury Screening

Background Non-contact lower limb injuries, including anterior cruciate ligament (ACL) tears and lateral ankle sprains, frequently occur during the sudden deceleration phase of single-leg landing. While laboratory force platforms represent the clinical gold standard for quantifying landing kinetics, the cost and spatial constraints restrict widespread on-field applications. Wearable inertial measurement units (IMUs) offer a portable alternative, but whether kinematic data from single-sensor configurations can accurately predict multi-planar impact loading and dynamic postural stability remains unclear. Methods Forty-eight recreationally active jumping athletes (24 males, 24 females; age: 22.4 ± 2.6 years; body mass: 64.8 ± 11.2 kg) performed standardized 30 cm single-leg drop landings (SLDL), as the baseline, and multi-planar single-leg drop landing (MSLDL) tasks. Kinematic features from a 9-sensor wearable IMU array (accelerometer, gyroscope, and combined modalities across 17 sensor configurations) were synchronized with force platform kinetic data. Nine outcome variables (body-weight normalized peak vertical Ground Reaction Force (vGRF), time to peak vGRF, vertical average/instantaneous loading rates (VALR, VILR)), dynamic postural stability indices for resultant, anteroposterior, and mediolateral directions (DPSI Total, DPSI AP, DPSI ML), and time-to-stabilization using the Ross Range method for ML and AP directions (TTS Fx and Fy). Six machine learning algorithms: ElasticNet, Support Vector Regression, Random Forest, LightGBM, XGBoost, and Multilayer Perceptron (MLP) Neural Networks were trained and evaluated using subject-isolated nested cross-validation, model optimization via grid-search, with Shapley Additive exPlanations (SHAP) analysis for feature importance decomposition. Results A total of 5508 ML regressors were built. ElasticNet yielded the highest predictive accuracy for VALR in SLDL ( R 2 = 0.815, nRMSE = 7.98%) and MSLDL ( R 2 = 0.744, nRMSE = 6.44%), alongside VILR in SLDL ( R 2 = 0.804), whereas gradient boosted trees (LightGBM) led MSLDL VILR ( R 2 = 0.788). Conversely, MLP Neural Networks dominated Peak vGRF/BW across both SLDL ( R 2 = 0.814, nRMSE = 6.61%) and MSLDL ( R 2 = 0.808, nRMSE = 6.72%), as well as dynamic postural stability (DPSI Total: R 2 = 0.641 to 0.655, nRMSE ≈ 4.1%). Sensor reduction demonstrated that a single shank placement retained over 90% of predictive accuracy retention vertical loading kinetics (Peak vGRF: R 2 = 0.730; VALR: R 2 = 0.740; VILR: R 2 = 0.770), while the thigh sensor enhanced the predictions in MSLDL loading rate. SHAP analysis demonstrated that early tibial shock kurtosis, non-landing limb jerk dynamics, and the landing shank-to-pelvis phase lag were key important kinematic features for estimating impact force and dynamic stability. Conclusions A single shank-mounted IMU paired with regularized linear or deep neural network regression model serves as a scientifically valid and field-deployable alternative to estimate laboratory force platforms metrics during SLDL and MSLDL conditions. Practitioners may consider applying the proper single- or 2-wearable IMU sensor placements as the monitoring or biofeedback tool for rehabilitation and injury prevention. Future studies should focus on the actual values of such IMU-based landing kinetic estimations on injury risk identification and prediction.

Authors

Institutions

Publication Details

Journal
Research Square
Published
2026-11-18
DOI
https://doi.org/10.21203/rs.3.rs-11431792/v1
Primary Topic
Knee injuries and reconstruction techniques
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Predicting Multi-planar Single-Leg Drop Landing Kinetic and Dynamic Stability Outcomes from Various Wearable Inertial Measurement Unit Configurations: A Machine Learning Approach with Implications for Field-Based Injury Screening

Indy Man Kit Ho, Lana Ružić, Daniel H. K. Chow, Jialiang Gu
1 links
Research Square
Knee injuries and reconstruction techniques
preprint

Predicting Multi-planar Single-Leg Drop Landing Kinetic and Dynamic Stability Outcomes from Various Wearable Inertial Measurement Unit Configurations: A Machine Learning Approach with Implications for Field-Based Injury Screening

Indy Man Kit Ho, Lana Ružić, Daniel H. K. Chow, Jialiang Gu
preprint en

Abstract

Background Non-contact lower limb injuries, including anterior cruciate ligament (ACL) tears and lateral ankle sprains, frequently occur during the sudden deceleration phase of single-leg landing. While laboratory force platforms represent the clinical gold standard for quantifying landing kinetics, the cost and spatial constraints restrict widespread on-field applications. Wearable inertial measurement units (IMUs) offer a portable alternative, but whether kinematic data from single-sensor configurations can accurately predict multi-planar impact loading and dynamic postural stability remains unclear. Methods Forty-eight recreationally active jumping athletes (24 males, 24 females; age: 22.4 ± 2.6 years; body mass: 64.8 ± 11.2 kg) performed standardized 30 cm single-leg drop landings (SLDL), as the baseline, and multi-planar single-leg drop landing (MSLDL) tasks. Kinematic features from a 9-sensor wearable IMU array (accelerometer, gyroscope, and combined modalities across 17 sensor configurations) were synchronized with force platform kinetic data. Nine outcome variables (body-weight normalized peak vertical Ground Reaction Force (vGRF), time to peak vGRF, vertical average/instantaneous loading rates (VALR, VILR)), dynamic postural stability indices for resultant, anteroposterior, and mediolateral directions (DPSI Total, DPSI AP, DPSI ML), and time-to-stabilization using the Ross Range method for ML and AP directions (TTS Fx and Fy). Six machine learning algorithms: ElasticNet, Support Vector Regression, Random Forest, LightGBM, XGBoost, and Multilayer Perceptron (MLP) Neural Networks were trained and evaluated using subject-isolated nested cross-validation, model optimization via grid-search, with Shapley Additive exPlanations (SHAP) analysis for feature importance decomposition. Results A total of 5508 ML regressors were built. ElasticNet yielded the highest predictive accuracy for VALR in SLDL ( R 2 = 0.815, nRMSE = 7.98%) and MSLDL ( R 2 = 0.744, nRMSE = 6.44%), alongside VILR in SLDL ( R 2 = 0.804), whereas gradient boosted trees (LightGBM) led MSLDL VILR ( R 2 = 0.788). Conversely, MLP Neural Networks dominated Peak vGRF/BW across both SLDL ( R 2 = 0.814, nRMSE = 6.61%) and MSLDL ( R 2 = 0.808, nRMSE = 6.72%), as well as dynamic postural stability (DPSI Total: R 2 = 0.641 to 0.655, nRMSE ≈ 4.1%). Sensor reduction demonstrated that a single shank placement retained over 90% of predictive accuracy retention vertical loading kinetics (Peak vGRF: R 2 = 0.730; VALR: R 2 = 0.740; VILR: R 2 = 0.770), while the thigh sensor enhanced the predictions in MSLDL loading rate. SHAP analysis demonstrated that early tibial shock kurtosis, non-landing limb jerk dynamics, and the landing shank-to-pelvis phase lag were key important kinematic features for estimating impact force and dynamic stability. Conclusions A single shank-mounted IMU paired with regularized linear or deep neural network regression model serves as a scientifically valid and field-deployable alternative to estimate laboratory force platforms metrics during SLDL and MSLDL conditions. Practitioners may consider applying the proper single- or 2-wearable IMU sensor placements as the monitoring or biofeedback tool for rehabilitation and injury prevention. Future studies should focus on the actual values of such IMU-based landing kinetic estimations on injury risk identification and prediction.

Research Square
University of Zagreb (HR), Education University of Hong Kong (HK), Technological and Higher Education Institute of Hong Kong (HK)
Good health and well-being
Knee injuries and reconstruction techniques
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon