Evaluating the performance of different machine learning algorithms based on SMOTE in predicting musculoskeletal disorders in elementary school students.

backgroundMusculoskeletal disorders (MSDs) are a major health concern for children. Traditional assessment methods, which are based on subjective assessments, may be inaccurate. The main objective of this research is to evaluate Synthetic Minority Over-sampling Technique(SMOTE)-based machine learning algorithms for predicting MSDs in elementary school students with an unbalanced dataset. This study is the first to use these algorithms to increase the accuracy of MSD prediction in this age group.

methodsThis cross-sectional study was conducted in 2024 on 438 primary school students (boys and girls, grades 1 to 6) in Hamedan, Iran. Random sampling was performed from 12 public and private schools. The dependent variable was the presence or absence of MSD, assessed using the Cornell questionnaire. Given the imbalanced nature of the data, SMOTE-based techniques were applied. Finally, the performance of six machine learning algorithms, including Random Forest (RF), Naive Bayes (NB), Artificial Neural Network (ANN), Decision Tree (DT), Extreme Gradient Boosting (XGBoost), and Support Vector Machine (SVM) was compared based on five SMOTE family techniques, including SVM-SMOTE, SMOTE-NC, ADASYN, Borderline SMOTE, and the SMOTE method, for predicting musculoskeletal disorders in primary school students.

resultsThis study analyzed 438 students, with 28 (6.39%) having musculoskeletal disorders (MSD). Machine learning models were tested on both imbalanced and SMOTE-balanced data. Before SMOTE, sensitivity was low (e.g., DT: 18%), but after applying SMOTE, it improved significantly (DT: 85%). Area Under the Curve (AUC) for RF and XGBoost increased from 65 to 99%. Among SMOTE-based techniques, SMOTE-NC achieved the highest accuracy for XGBoost (93.65%), RF (93.41%), ANN (92.80%), SVM (90.24%), and DT (86.58%). Borderline-SMOTE yielded the highest accuracy (96%) for XGBoost and RF. RF and XGBoost performed best overall. Key predictors of MSD were regional facilities, BMI, and gender.

conclusionThe findings of this study demonstrate the effectiveness of SMOTE in addressing class imbalance and improving the accuracy of MSD prediction among students. Overall, RF and XGBoost outperformed other algorithms. Additionally, key factors influencing MSD included regional facilities, body mass index (BMI), and gender, highlighting the significance of environmental and physiological factors in the development of this disorder.