Comparative analysis of machine learning models for efficient low back pain prediction using demographic and lifestyle factors.

Abstract

Background: Low back pain (LBP) is one of the most frequently occurring musculoskeletal disorders, and factors such as lifestyle as well as individual characteristics are associated with LBP.

Objective: The purpose of this study was to develop and compare efficient low back pain prediction models using easily obtainable demographic and lifestyle factors.

Methods: Data from adult men and women aged 50 years or older collected from the Korean National Health and Nutrition Examination Survey (KNHANES) were used. The dataset included 22 predictor variables, including demographic, physical activity, occupational, and lifestyle factors. Four machine learning algorithms, including XGBoost, LGBM, CatBoost, and RandomForest, were used to develop predictive models.

Results: All models achieved an accuracy greater than 0.8, with the LGBM model outperforming the others with an accuracy of 0.830. The CatBoost model had the highest sensitivity (0.804), while the LGBM model showed the highest specificity (0.884) and F1-Score (0.821). Feature importance analysis revealed that EQ-5D was the most critical variable across all models.

Conclusion: In this study, an efficient LBP prediction model was developed using easily accessible variables. Using this model, it may be helpful to identify the risk of LBP in advance or establish prevention strategies in subjects who have difficulty accessing medical facilities.