Table 3 Most important predictors in five machine learning models
Performing model | Most important predictors | |
|---|---|---|
Feature permutation importance | Shapley Additive Explanations | |
Random Forest | Ankle dorsiflexion ROM, strength ratio of evertor in a neutral position to invertor, age, lunge angle, BMI, YBT | Low ankle dorsiflexion ROM, high BMI, low lunge angle, low strength ratio of evertor in a neutral position to invertor, old age, high RCSP |
Extreme Gradient Boosting | Age, ankle dorsiflexion ROM, strength ratio of evertor in plantar flexion to invertor, lunge angle, BMI, strength ratio of evertor in a neutral position to invertor | Low ankle dorsiflexion ROM, old age, low strength ratio of evertor in plantar flexion to invertor, low strength ratio of evertor in a neutral position to invertor, low lunge angle, high BMI |
Logistic Regression | Ankle dorsiflexion ROM, success of the ECSLS, RCSP, ankle dorsiflexor strength, lunge angle, a number of balance retrials of the ECSLS | Fail of the ECSLS, low ankle dorsiflexion ROM, high RCSP, low ankle dorsiflexor strength, low lunge angle, a high number of balance retrials of the ECSLS |
Support Vector Machine | Success of the ECSLS, ankle dorsiflexion ROM, RCSP, the number of balance retrials of the ECSLS, age, strength ratio of evertor in a neutral position to invertor | Fail of the ECSLS, low ankle dorsiflexion ROM, a high number of balance retrials of the ECSLS, low strength ratio of evertor in a neutral position to invertor, low lunge angle, high RCSP |
Naïve Bayes | Ankle dorsiflexion ROM, RCSP, age, strength ratio of evertor in plantar flexion to invertor, BMI, ankle dorsiflexor strength | Low ankle dorsiflexion ROM, old age, high BMI, long work duration, a high number of balance retrials of the ECSLS, high ankle dorsiflexor strength |