Page 63 - THEi Student Applied Research Presentations 2024
P. 63

MACHINE LEARNING-BASED PREDICTION OF
    RUNNING INJURIES: POOLED ANALYSIS OF

    MOTIONMETRIX DATA AND SCREENING TESTS


    ABSTRACT               Poster


    RESEARCH BACKGROUND

    Running-related  injuries,  particularly  overuse  injuries,  are  significant  concerns  for
    runners.  Relying  solely  on  traditional  measures  like  running  volume  oversimplifies
    the assessment of training stress, often neglecting important factors such as ground
    reaction force and foot-strike pattern. Instead of the linear and unidirectional causality
    view of sports injury etiology, the complex system perspective proposed a multifactorial
    nature of injuries, emphasizing unknown interactions and varying weights among   LI MEI HUNG
    features.
                                                                                 BSocSc (Hons) in Sports and
    METHODOLOGY                                                                  Recreation Management
    A  total  of  59  well-trained  long-distance  runners  were  recruited.  Demographic   Department of Sport and Recreation
    information, running behavior, and injury history were collected through the survey.
    Participants completed a 1-minute MotionMetrix running assessment at 12 km/h with
    0% inclination, and three screening tests, including the knee-to-wall lunge test, hip
    abduction strength test, and single-leg squat test. One hot encoding and oversampling
    techniques were employed for data preprocessing. Machine learning models, including   OBJECTIVES
    random forest, support vector machine, and extreme gradient boost, were implemented
    in Python. The dataset was split into 80% for training and 20% for testing, and k-fold   The current study integrated survey
    cross-validation was utilized.                                               data, screening tests, and the joint
                                                                                 kinematics  and  kinetics  results  from
    FINDINGS                                                                     the  MotionMetrix  markerless  running
    The XGBoost model exhibited the highest evaluation of accuracy (0.717), F-1 score   assessment system, to identify the
    (0.759), and AUC (0.735), which suggested its great accuracy and superior predictive   high-importance features, and develop
    and discriminative capabilities in identifying runners at risk of injury. The most   a machine learning-based predictive
    important features included weekly running distance, hip abduction strength, and   model for assessing running injury risk.
    running experience.

    ABOUT THE INVESTIGATOR


    Hey, I'm LI Mei Hung Parker, a Year-4 sports therapy student at SRM. Studying can be
    enjoyable, but if it's not your thing, it's important to find something that truly excites
    you. Lately, I've been embracing traveling and exploring life’s unpredictability. One of
    my goals is to become a skilled and knowledgeable sports therapist. I'm passionate
    about helping people have a blast while staying active and living their best, healthiest
    lives.
    Here's my super supervisor: Mr. Indy HO!





















 Student Applied Research Presentations 2024                                  Student Applied Research Presentations 2024  63
   58   59   60   61   62   63   64   65   66   67   68