Table 3.
Average testing accuracy achieved with two classical machine learning (ML) methods, and three DL methods, for six sensors combinations. LOSO evaluation (%). A—accelerometer, G—gyroscope, M—magnetometer, R—rotation vector *.
| Method | Sensor Combination | |||||
|---|---|---|---|---|---|---|
| A | G | AG | GR | AGM | AGR | |
| RF | 82.8 | 81.6 | 83.5 | 82.4 | 83.4 | 84.1 |
| SVM | 81.3 | 81.6 | 81.8 | 81.2 | 82.2 | 84.0 |
| CNN | 77.5 | 79.1 | 80.9 | 77.5 | 76.8 | 85.2 |
| LSTM | 83.0 | 83.4 | 86.1 | 81.7 | 79.8 | 86.4 |
| Proposed | 83.7 | 85.8 | 87.8 | 85.1 | 88.3 | 88.9 |
* The results for the random forest (RF) and support vector machine (SVM) were achieved with manually extracted features, and CNN, LSTM, and the proposed method work with raw signals.