Search

In this paper, four tri-axis accelerometers are used to collect mechanomyography (MMG) signal of four thigh muscles during gait movements. Three time-domain features, two frequency-domain features, and 64 time-frequency domain features of the MMG were extracted, and the features were ranked by feature importance using random forest (RF). In addition, RF and a support vector machine (SVM) are utilized as classifiers, and the relationship between the number of features and the recognition accuracy is analyzed. After the feature importance analysis, the principal component analysis (PCA) was used to perform the dimensionality reduction, and the relationship between the reduced dimension and the accuracy was studied, moreover, RF and SVM were compared. The conclusion in the paper can help to select the features that have the greatest contribution to the gait recognition: zero-crossing rate, mean power frequency, median frequency, and energy coefficients of wavelet packet, providing theoretical basis for researchers to select suitable features in the future.