فهرست نشریات علمی دسترسی آزاد ایران

Injury prevention is a critical concern in sports education, where students often face elevated risks due to intensive training and lifestyle factors. Traditional decision-tree models, while interpretable, are limited in their ability to handle incomplete or contradictory data, as they typically rely on imputation strategies that may introduce bias and distort the distribution of observations. This study addresses this gap by proposing the IndetermSoft–C4.5 algorithm, an extension of the classical C4.5 decision tree that explicitly incorporates indeterminate values. Rather than discarding ambiguous records, the algorithm distributes them across feasible categories using fractional weights, thereby preserving uncertainty throughout the learning process. The research is based on a dataset of 245 sports pedagogy students, including demographic, anthropometric, and training-related attributes. Model performance was assessed using stratified cross-validation and benchmarked against classical C4.5 with imputation, Logistic Regression (LF), Random Forest (RF), and Gradient Boosting (GB). Results demonstrate that the IndetermSoft–C4.5 approach achieved an accuracy of 86.1% and an area under the curve of 0.85, outperforming the classical baseline while remaining competitive with ensemble methods. The model maintained a balance between sensitivity and specificity, generated interpretable decision rules, and showed robustness under varying hyperparameter configurations. These findings highlight the potential of IndetermSoft–C4.5 as a reliable and transparent predictive framework for sports injury risk, with broader implications for uncertainty-aware decision-making in educational and health domains.