Akademik Veri Yönetim Sistemi
INTERNATIONAL KAYSERI SCIENTIFIC RESEARCH AND INNOVATION CONGRESS , Kayseri, Türkiye, 30 - 31 Mayıs 2026, ss.230-242, (Tam Metin Bildiri)
This study aims to model and predict injury risk in athletes using machine learning methods based on two different open datasets. The Personalized Sports Health dataset includes variables related to body composition, training load, cardiovascular indicators, and recovery parameters, while the Biomechanical Analysis for Injury Prevention dataset covers biomechanical measurements such as muscle activation, joint loading, joint angles, and movement symmetry. Each dataset was modeled separately according to its own measurement context. Random Forest, Support Vector Machine (SVM), XGBoost, and Artificial Neural Network (MLP) algorithms were applied to both datasets. The datasets were not merged at the raw-data level; instead, the modeling results were compared through common conceptual feature groups such as body composition, cardiovascular indicators, muscle activation, and joint loading. In this way, significant feature groups were identified across the two datasets. In the Personalized dataset, the best performance was achieved with the MLP/ANN model, whereas in the Biomechanical dataset, the XGBoost algorithm produced the best results. Variable importance analysis was conducted for the grouping process, revealing that recovery/lifestyle and injury history variable groups were the mosT influential factors in the Personalized dataset, while joint loading and muscle activation groups were the most significant in the Biomechanical dataset. The findings indicate that injury risk has a multidimensional structure and support the need for developing holistic assessment systems.