JMIR Publications

ABSTRACT

Background:

For rehabilitation training systems, it is essential to automatically record and recognize exercise motions, especially when more than one type of exercises are performed without predefined sequences. Most of the motion recognition methods are based on feature engineering and machine learning algorithms. Time-domain and frequency-domain features are extracted from original time series collected by sensor nodes. For high-dimensional data, feature selection plays an important role in improving the performance of motion recognition. Existing feature selection methods can be categorized into filter and wrapper methods. Wrapper methods usually achieve better performances than filter methods, however, in most cases, they are computationally intensive and the feature subset obtained is usually optimized only for the specific learning algorithm.

Objective:

This study aims to provide a feature selection method for motion recognition of upper-limb rehabilitation exercises, to improve the recognition accuracy and time efficiency.

Methods:

Motion data of five types of upper-limb rehabilitation exercises from five subjects are collected by customized inertial measurement unit (IMU) node. A total of 60 time-domain and frequency-domain features are extracted from the original motion data. A hybrid feature selection by combining filter and wrapper methods (FESCOM) is proposed to eliminate irrelevant features for motion recognition of rehabilitation exercises. In the filter stage, candidate features are first selected from the original feature set according to the significance for motion recognition. Then, three classification algorithms including k-Nearest Neighbors (kNN), Naïve Bayes (NB) and Random Forest (RF) are evaluated as the wrapping component to further refine the features from a candidate feature set. The performance of the proposed FESCOM method is verified by experiments on motion recognition of upper-limb rehabilitation exercises and compared with the traditional wrapper method.

Results:

Using kNN, NB and RF as the wrapping components, the classification error rate of the proposed FESCOM method is 2.78%, 074% and 4.63%, respectively, and the time consumed on feature selection in each iteration is 5.63s, 55.38s, and 277.03s, respectively.

Conclusions:

Experiment results demonstrate that the proposed FESCOM method achieves better performance than traditional wrapper method in terms of recognition accuracy and time efficiency. This approach serves to improve feature selection and classification algorithm selection for upper-limb motion recognition based on wearable sensor data, which can be extended to exercise assessment in autonomous rehabilitation systems.