Automated Detection of Lower Back Pain Using Machine Learning and SMOTE-Based Data Augmentation

Low Back Pain (LBP) is a leading global health concern, affecting up to 80% of individuals at some point in their lives and ranking among the most common causes of chronic disability and work absenteeism. Despite advancements in treatment, accurate and scalable diagnostic tools remain limited. Traditional diagnostic methods rely heavily on clinical expertise and imaging, which are often time-consuming, subjective, and inaccessible in resource-limited settings. Recent literature underscores the potential of Machine Learning (ML) for automating LBP detection, but challenges such as imbalanced datasets and insufficient model generalizability persist. This study introduces a robust ML pipeline for automatic LBP classification using data from the publicly available international dataset - Kaggle. The workflow incorporates data type normalization, outlier elimination, and feature distribution analysis, followed by class rebalancing through the Synthetic Minority Oversampling Technique (SMOTE). Three ML classifiers-Decision Tree (DT), Support Vector Machine (SVM), and Artificial Neural Network (ANN)-are trained and evaluated on both imbalanced and SMOTE-balanced datasets. Experimental results demonstrate a significant boost in classification performance post-balancing, with the ANN model achieving the highest accuracy (96.43%) and F1-score (96.47%). This work confirms that integrating effective preprocessing with optimized model selection can deliver accurate, scalable, and automated LBP detection-offering a meaningful step toward smarter musculoskeletal diagnostics.