Comparison of Machine Learning Algorithms for Patient Handling Recognition based on Body Mechanics

Abstract

Caregivers have low back pains due to patient handling. Therefore, we have been developing a monitoring system for patient handling designed to prevent low back pains. In this study, we propose a recognition method for patient handling using inertial measurement unit (IMU) and insole pressure sensor for a wearable monitoring system. Our proposed method recognizes recommended body mechanics-based patient handling by machine learning with trunk angle and ground reaction force on feet. Accuracy of the proposed method was evaluated by the experiment in a laboratory environment. Furthermore, accuracies of 7 common algorithms for machine learning are compared for consideration of the most suitable algorithm for the proposed method. Selected 7 algorithms were artificial neural network (ANN), decision tree (DT), k-nearest neighbor (KNN), logistic, random forest (RF), naive Bayes (NB), and support vector machine (SVM). The study participants were 5 young males. Each participant performed a stand assist motion with 2 conditions (natural and body mechanics) 10 trials for each condition. Accuracies were calculated by 10-fold cross validation with 100 data. The results suggested that our proposed method could recognize 2 stand assist motions with high accuracy more than 0.9 of the time in 3 machine learning algorithms. Furthermore, the best algorithm (RF) could recognize 2 stand assist motions with 0.97 accuracy. These results indicated that the proposed method could recognize patient handling motion and could be applied to a monitoring system to prevent low back pain among caregivers. In addition, this study found the best algorithm for patient handling recognition.