Factors contributing to chronic ankle instability in parcel delivery workers based on machine learning techniques.

Abstract

Background: Ankle injuries in parcel delivery workers (PDWs) are most often caused by trips. Ankle sprains have high recurrence rates and are associated with chronic ankle instability (CAI). This study aimed to develop, determine, and compare the predictive performance of statistical machine learning models to classify PDWs with and without CAI using postural control, ankle range of motion, ankle joint muscle strength, and anatomical deformity variables.

Methods: 244 PDWs who had worked in parcel delivery for more than 6 months were screened for eligibility. Thirteen predictors were included in the study: 12 numeric (age, body mass index, work duration, the number of balance retrials eyes-closed single-limb stance, Y-balance test, ankle dorsiflexion range of motion, lunge angle, strength ratio of the evertor in plantar flexion and neutral position to the invertor, ankle dorsiflexor strength, navicular drop, and resting calcaneal stance position) and one categorical (success of the eyes-closed single-limb stance). Five machine learning algorithms, including LASSO logistic regression, Extreme Gradient boosting machine, support vector machine, Naïve Bayes machine, and random forest-were trained.

Results: The support vector machine and random forest models confirmed good predictive performance in the training and test datasets, respectively, for PDWs. For the Shapley Additive Explanations, among the five machine learning models, the variables entered into three or more models were low ankle dorsiflexion range of motion, low lunge angle, high body mass index, old age, a high number of balance retrials of the eyes-closed single-limb stance, and low strength ratio of the evertor in the neutral position to the invertor.

Conclusion: Our approach produced machine learning models to classify PDWs with and without CAI and confirmed good predictive performance in PDWs.