Table 7.
This table presents 3 sets of features used as inputs to the described machine learning classification algorithms. The first set includes features extracted from the data of the insoles, the second includes features extracted from the data of the IMU-based system, and the third includes a subset of the combination of the previous two sets. In the third set the designation “ins.” refers to the insole-based system, while “IMU” to the IMU-based system. Those sets are presented in decreasing order of importance from top to bottom. The feature importance values were calculated using the Shapley Additive Explanations (SHAP) framework.
| Insole Features | IMU Features | Insole & IMU Features | |
|---|---|---|---|
| F01 | Right loading response (%, std) | Accel. max energy (asym., std) | Right load. response (%, std, ins.) |
| F02 | Left loading response (%, mean) | Walking cadence (asym.) | Left pre-swing (mean, ins.) |
| F03 | Pre-swing variability | Range of shank’s motion | Total accel. energy (asym., IMU) |
| F04 | Left loading response (%, std) | Total gyroscope’s energy | Total accelerometer’s energy (IMU) |
| F05 | Left pre-swing (mean) | Normalized stride length | Pre-swing variability (ins.) |
| F06 | Right loading response (std) | Cumul. accelerometer’s energy | Normalized stride length (IMU) |
| F07 | Left loading response (std) | Norm. walking speed (asym.) | Right load. response (std, ins.) |
| F08 | Right terminal stance (%, mean) | Maximum rotation | Left load. response (std, ins.) |
| F09 | Left single support (std) | Number of steps (count) | Range of shank’s motion (IMU) |
| F10 | Right terminal stance (mean) | Contact time (asym.) | Norm. walk. speed (asym., IMU) |