Table 3.

Improvements in this study compared to previous studies.

Study	Badura’s Study	Kim’s Study	This Study
Classification model	Multi-layer perceptron (MLP)	Support vector Machine (SVM)	Double head 1D-CNN and single head GRU stacking ensemble
Feature extraction	Manual (Frequency and time domain feature, Feature selection: Fisher’s linear discriminant)	Manual (Frequency domain and energy feature, Feature selection: KPCA)	Automatic in deep learning
Sampling rate of data (Hz)	100	100	20 (Introduce data down-sampling)
Data imbalance problem	Yes	Yes	No (Introduce data augmentation)
Amount of experimental data	63	53	78
Evaluation method	Random split Training: Test = 7:3	Random split Training: Test = 7:3	Mean accuracy of 10-fold cross validation