Table 3. Study results characteristics.
Author (Year) ML Algorithm Used Best- Performing Algorithm Model Performance Model Interpretability (Important Injury Predictors)
Lopez-Valenciano et al. (2018) C4.5, SimpleCart, ADTree, RF ADTree AUC = 0.75, Sensitivity = 0.66, Specificity = 0.69 Interpretable by Design (sport devaluation, history of muscle injury in last season)
Ruddy et al. (2018) Naïve Bayes, LR, RF, SVM, NN Naïve Bayes AUC = 0.60 Not Reported
Ayala et al. (2019) J48, SimpleCart, ADTree ADTree AUC = 0.84, Sensitivity = 0.78, Specificity = 0.84 Interpretable by Design (sleep quality, history of HSI last season, range of motion – passive hip flexion with knee extended)
Connaboy et al. (2019) CHAID DT AUC = 0.91 Interpretable by Design (knee flexion angle asymmetry, body mass)
Henriquez et al. (2020) RF RF AUC = 0.69 Interpretable by Design (hip external rotation strength, hip adductor strength, straight leg raises)
Oliver et al. (2020) LR, DT DT AUC = 0.66, Sensitivity = 0.56, Specificity = 0.74 Interpretable by Design (single leg counter movement jump peak vertical ground reaction force asymmetry, body mass, leg length)
Jauhiainen et al. (2021) RF, LR, SVM LR AUC = 0.65 Interpretable by Design (sex, body mass index, hamstring flexion non-dominant, KT1000 dominant)
Ruiz-Perez et al. (2021) C4.5, ADTree, KNN, SVM SVM AUC = 0.77, Sensitivity = 0.66, Specificity = 0.62 Interpretable by Design (hip flexion ROM, ankle dorsiflexion ROM)
Bogaert et al. (2022) LR, RF, SVM SVM Male (AUC = 0.62), Female (AUC = 0.65) Logistic Regression (Male: vertical acceleration-derived features; Female: medial-lateral-acceleration-derived features)
Jauhiainen et al. (2022) RF, LR, SVM SVM AUC = 0.63 Not Reported
Huang et al. (2022) dFusionModel dFusionModel Precision = 0.93, Sensitivity = 0.92 SHAP (Minimal LENCI: stress, squat 1RM; Mild LENCI: sRPE, sleep, urine protein, urine blood)
Lu et al. (2022) Elastic Net, RF, XGBoost, SVM, NN, LR XGBoost AUC = 0.84 SHAP (history of a back, quadriceps, hamstring, groin, or ankle injury; Concussion within the previous 8 weeks; Total count of previous injuries.)
Huang et al. (2023) Cost-NN, LR, RF, XGBoost Cost-NN AUC = 0.86, Precision = 0.64, Sensitivity = 0.87 SHAP (hexagon agility test, three-quarter court sprint)
Javier Robles-Palazon et al. (2023) C4.5, ADTree, SVM, KNN SVM AUC = 0.70, Sensitivity = 0.54, Specificity = 0.74 SHAP (knee maximum displacement (dominant leg) in the drop vertical jump, landing bilateral peak vertical ground reaction force (single-leg countermovement jump), BMI)
Kolodziej et al. (2023) LASSO LR LASSO LR AUC = 0.63, Sensitivity = 0.35, Specificity = 0.79 Interpretable by Design (concentric knee extensor peak torque, hip transversal plane moment in the SLDL, COP sway)
ADTree, alternating decision tree; RF, random tree; LR, logistic regression; SVM, support vector machine; NN, neural network; CHAID, chi-square automatic interaction detection; KNN, k-nearest neighbor; XGBoost, extreme gradient boosting; dFusionModel, RF-based fusion of XGBoost submodels; Cost-NN, cost-sensitive neural network.