Development of IMU Sensor-Based XGBoost Model for Patients with Elbow Joint Damage

Abstract

Elbow joint damage requires rehabilitation through continuous passive motion (CPM) devices or additional surgery, underscoring the significance of effective and consistent post-surgical rehabilitation. In this paper, we present an IMU sensor-based XGBoost model specifically designed for patients with elbow joint impairments, aiming to overcome the limitations of existing elbow rehabilitation methods. The proposed XGBoost model employs Softmax as its objective function to facilitate multi-class classification, while utilizing an Error function to minimize the error rate. For model training, we compiled a dataset using subjects, adhering to government guidelines for recommended elbow exercise postures and angles. We evaluated the performance of the implemented model by conducting comparisons with the tree-based ensemble model Random Forest, and the commercial model Teachable Machine. The results of this comparison showed that our proposed model achieved an average classification performance of 93.8%, while the Random Forest model exhibited an average of 91.6%, and Teachable Machine demonstrated an average of 94.9%. In conclusion, the performance of our implemented model was found to be comparable to that of existing commercial models. Moving forward, we plan to broaden our research scope by incorporating electromyography (EMG) signals, with the intention of applying our findings to various applications such as lower limb rehabilitation, prosthetic manipulation, and elbow Continuous Passive Motion (CPM) rehabilitation.