Machine Learning Approach for Fatigue Estimation in Sit-to-Stand Exercise

Abstract

Physical exercise (PE) has become an essential tool for different rehabilitation programs. High-intensity exercises (HIEs) have been demonstrated to provide better results in general health conditions, compared with low and moderate-intensity exercises. In this context, monitoring of a patients' condition is essential to avoid extreme fatigue conditions, which may cause physical and physiological complications. Different methods have been proposed for fatigue estimation, such as: monitoring the subject's physiological parameters and subjective scales. However, there is still a need for practical procedures that provide an objective estimation, especially for HIEs. In this work, considering that the sit-to-stand (STS) exercise is one of the most implemented in physical rehabilitation, a computational model for estimating fatigue during this exercise is proposed. A study with 60 healthy volunteers was carried out to obtain a data set to develop and evaluate the proposed model. According to the literature, this model estimates three fatigue conditions (low, moderate, and high) by monitoring 32 STS kinematic features and the heart rate from a set of ambulatory sensors (Kinect and Zephyr sensors). Results show that a random forest model composed of 60 sub-classifiers presented an accuracy of 82.5% in the classification task. Moreover, results suggest that the movement of the upper body part is the most relevant feature for fatigue estimation. Movements of the lower body and the heart rate also contribute to essential information for identifying the fatigue condition. This work presents a promising tool for physical rehabilitation.

Keywords: Kinect; fatigue estimation; machine learning; physical exercise; physical rehabilitation; sit-to-stand.

Figure 1

Borg CR10 definition table.

Figure 2

Set-up of the study and sit-to-stand representation, (A) standing position and (B) sitting position.

Figure 3

Example data of one volunteer test, (A) M_hip vertical signal, (B) heart rate signal and (C) Borg CR10 values.

Figure 4

M_hip vertical movement signal, maximum, minimum and phase detection.

Figure 5

Features extraction process from the Kinect row data, (A) M_hip vertical movement signal, (B) M_hip vertical velocity signal, (C) nee flexo-extension signal and (D) Knee flexo-extension velocity signal.

Figure 6

Linear interpolation of the Borg values every 10 s.

Figure 7

Selection of the five nearest sit-to-stand cycles to each Borg Value.

Figure 8

Selection of the five nearest heart rate records to each Borg value.

Figure 9

The behavior of the features normalized and fatigue level example, (A) Borg CR10 interpolated, (B) stand-to-stand time normalized, (C) Knee flexo-extension max velocity normalized, (D) Hip flexo-extension range normalized and (E) Heart rate normalized.

Figure 10

Data set representation composed of 660 STS registers, 33 features and the fatigue target.

Figure 11

Cross validation process for the machine learning model training and assessment.

Figure 12

Mean and standard deviation values of each feature, according to the 3 fatigue conditions. (A) corresponds to the features values between 1 to 11; (B) corresponds to the features values among 12 to 22; (C) corresponds to the features values between 23 to 33.

Figure 13

Features scatter graphs regarding the stand-to-stand time, (A) stand-to-stand time vs. sit-to-stand time, (B) sit-to-stand time vs. heart rate, (C) sit-to-stand time range vs. M_shoulder depth range, and (D) sit-to-stand time vs. M_hip max depth velocity.

Figure 14

Distribution diagram obtained from applying the Uniform Manifold Approximation and Projection technique.

Figure 15

Confusion matrix of (a) RF classifier, (b) SVM classifier, (c) ANN classifier, (d) LR classifier, and (e) KNN classifier.

Figure 16

Box plot of the performance metric results for the five best machine learning methods.

Figure 17

STS physical exercise intervals vs. fatigue conditions on the test set. The thinnest dots represent the ground-truth values, and the thickest dots are fatigue predictions.

Figure 18

Gender comparison of STS physical exercise intervals versus fatigue condition on the test set. Crosses and × symbols describe the ground-truth values, and the thickest dots present; (a) represents the fatigue condition in STS physical exercise intervals in male patients and (b) represents the fatigue condition in STS physical exercise intervals in female patients.

Figure 19

Relative importance of Features for the random forest model.