Comparative study of machine learning models to predict mobility in patients with low back pain using inertial sensors

Low back pain is one of the most common causes of disability and considerably affects life by impairing the quality of life of those who experience it. The purpose of the study is to compare the mobility of five machine learning models (LightGBM, XGBoost, HistGradientBoostingRegressor, GradientBoostin- gRegressor, and StackingRegressor) to predict mobility in patients with low back pain attended by inertial sensors. The MoCap database offered 2160 simulated patient samples that implemented three functional movements: flexion-extension, rotation, and lateralization. Synthetic data, normalization, and correlation analysis were applied as part of the preprocessing. The models were analyzed according to the MAE, MSE, and R2 parameters and performed on a 5-fold cross-validation. Additionally, to identify the statistically significant difference in the models, ANOVA and Tukey HSD tests were performed. The GradientBoostin- gRegressor was found to perform significantly better with flexion-extension and lateralization movements; however, its rotation is negligible. The study demonstrates the potential of combining affordable sensor technologies with advanced machine learning techniques to democratize access to highly accurate diag- nostic and monitoring tools in clinical and home settings. Furthermore, it can be adopted in disease rehabilitation as a noninvasive tool to facilitate patient follow-up, optimize individual therapy plans, and reduce the socioeconomic costs of back pain, offering a scalable and cost-effective solution for mobility monitoring, which would directly benefit both clinicians and patients.

• Auteur principal: Carlosama Quinatoa, Jeremy Fabrizio (author)
• Format: bachelorThesis
• Publié: 2025
• Sujets: Aprendizaje automático; Tecnología portátil; Modelos de predicción; Machine learning; Wearable technology; Clinical evaluation
• Accès en ligne: https://repositorio.yachaytech.edu.ec/handle/123456789/1034