Abstract details
| Abstract-ID: | 1250 |
| Title of the paper: | Predicting the running injuries with kinematics and kinetics data of MotionMetrix using Machine Learning Algorithms |
| Authors: | Li, M.H., Luk, J.T.C., LAM, S.S., LAW, V.K.H., LEE, G., Ho, I.M.K. |
| Institution: | Technological and Higher Education Institute of Hong Kong |
| Department: | Faculty of Management and Hospitality |
| Country: | Hong Kong |
| Abstract text | INTRODUCTION: Running-related injuries are a significant concern among runners, particularly overuse injuries, impacting their performance and recovery time. Despite efforts to prevent these injuries, current research and programs have shown limited success in reducing injury rates and recurrence. Traditional measures such as solely relying on running volume may provide an oversimplified view of training stress and other factors like ground reaction force and foot-strike pattern are often overlooked. Instead of the linear and unidirectional causality view of sports injury etiology, the complex system perspective proposed a multifactorial unknown interaction among the web of determinants with different weights. The current study combined both clinical tests, survey data, and the joint kinematics and kinetics results generated by the novel markerless running assessment system, MotionMetrix for performing predictive analytics on running injury risk. METHODS: A total of 30 trained long-distance runners participated in the running assessment. Participants completed the 1-minute MotionMetrix test on a motorized treadmill using 11 km/h speed and 0% inclination. Meanwhile, the clinical tests including the knee-to-wall ankle dorsiflexion mobility test, hip abduction strength test, single-leg squat test, and the foot posture index, as well as the short survey regarding the training and running injury history were recorded. The occurrence of any running injury was the outcome variable while others were treated as the predictors. One-hot-encoding was used to pre-process the nominal categorical data such as gender while the redundant or irrelevant variables were eliminated or filtered out to reduce the dimensionality. Machine learning models were produced using the library Scikit Learn in Python including random forest (RF), support vector machine (SVM), lightGBM (LGBM), and Neural Network (NN). Training and testing data were obtained by 80% and 20% of the original dataset respectively while the k-fold cross-validations were used to reduce the overfitting issues. RESULTS: The model accuracy and F1 scores were compared to select the best-performed model. Furthermore, the feature importance or SHAP values were calculated to identify the most influential predictors. CONCLUSION: The combined use of the markerless 3D running gait analysis tool, MotionMetrix system, survey data of training volume, and clinical tests were able to predict the running injuries using the selected machine learning algorithms. It offers a feasible, efficient, and cost-effective alternative overcoming the limitations of traditional laboratory-based evaluation methods. Running coaches can make a better decision and training programs with data analytics using machine learning on a larger dataset. |
| Topic: | Biomechanics |
| Keyword I: | |
| Keyword II: | |
| Keyword III: | |