Comparison of the Accuracy of Ground Reaction Force Component Estimation between Supervised Machine Learning and Deep Learning Methods Using Pressure Insoles
Abstract
:1. Introduction
2. Materials and Methods
2.1. Materials and Protocol
- (1)
- normal walk of 6 to 10 trials, resulting in (7–20 steps on the force plates) (Figure 3);
- (2)
- slow walk of 6 to 10 trials, resulting in (8–22 steps on the force plates) (Figure 3);
- (3)
- static situation (standing still) (2–5 s) (Figure 4);
- (4)
- static situation carrying a 5 kg load (static situation with CL) (2–5 s) (Figure 4);
- (5)
- carrying a 5 kg load from bottom to top and vice versa (bottom-top with CL) (7–18 trials) (Figure 5);
- (6)
- carrying a 5 kg load from left to right and vice versa (left-right with CL) (5–11 trials) (Figure 6).
2.2. Data Preprocessing
- Linear interpolation of missing data of the insole
- Time-shift synchronization between the insole and force plate data
- Deletion of the data outside the force plate for the two walking activities
2.3. Determination of the Optimal Architecture and Parameters of SML and DL Methods for GRF Component Estimation
- Artificial Neural Network (ANN)
- (1)
- initialization: 1 input layer with 16 neurons, 2 hidden layers of (256, 128) neurons, activation function: sigmoid, normalization method for input (PP) and output (GRF components) data: mean, optimizer: Adamax, batch size: 32, learning rate: 0.01;
- (2)
- modify the optimizer: Adagrad, AdamW, Adadelta, Adam, Adamax, NAdam, RMSprop, Stochastic Gradient Descent (SGD) with different momentum values: 0, 0.5, and 0.9;
- (3)
- modify the learning rate: 0.04, 0.08, 0.01, 0.005, and 0.008;
- (4)
- modify the batch size: 1, 4, 16, 32, 64, 128, and 256;
- (5)
- modify the number of hidden layers (and their neurons): 1 layer: (150); 2 layers: (50, 50), (125, 125), (256, 128), and (128, 256); 3 layers: (256, 256, 128); 4 layers: (100, 100, 100, 100);
- (6)
- modify the activation function: tanh, leaky relu, softSign, relu, sigmoïde, wavelet, softPlus, and elu;
- (7)
- modify the normalization method: Min-Max in the range [0, 1] and [−1, 1], Mean, Z-Score, Robust Scaler, Vector Standardization, Maximum Linear Standardization, Decimal Scaling, Median, Tanh, Body Weight (BW), and Length Insole. These 12 normalization methods are explained in this study [24].
- Long Short-Term Memory (LSTM)
- (1)
- initialization: 1 input layer with 16 neurons, 1 LSTM layer with 128 cells, an ANN with 2 hidden layers of (128, 50) neurons, an input sequence size equal to 20 (the LSTM uses the PP data at the current time t as well as previous samples at times t − 19, …, t − 1 to estimate GRF components at the time t);
- (2)
- modify the number of LSTM layers (and their cells): 1 LSTM layer (128); 2 LSTM layers: (400, 200), (400, 100), and (800, 400);
- (3)
- modify the number of hidden layers of ANN (and their neurons): 1 layer (400); 2 layers (256, 128); 3 layers: (400, 100, 50) and (400, 300, 150);
- (4)
- test BLSTM instead of LSTM with the best architecture and parameters obtained from these 3 steps;
- (5)
- modify the number of BLSTM layers (and their cell): 1 layer (400); 2 layers: (400, 200) and (600, 200);
- (6)
- modify the size of the input sequence: 20, 40, 60, 80, and 100.
- Convolutional Neural Network (CNN)
- (1)
- initialization: 1 input layer with 16 neurons, 3 one-dimensional convolutional layers (Conv1D): Conv1D (number of filters = 14, kernel size = 4), Conv1D (number of filters = 10, kernel size = 4), Conv1D (number of filters = 8, kernel size = 4), 3 pooling layers: AvgPool1d (kernel size = 2), an ANN with 1 hidden layer of (32) neurons;
- (2)
- test MaxPool1d instead of AvgPool1d;
- (3)
- modify the kernel size of the convolutional layers (Conv1D): 2, 3, 4, and 5;
- (4)
- modify the number of convolutional layers: 1, 2, and 3;
- (5)
- modify the number of filters of Conv1D: 4, 8, 14, and 20;
- (6)
- modify the number of hidden layers of ANN (and their neurons): 1 layer (96); 2 layers (96, 50); 3 layers: (96, 60, 25) and (96, 100, 50).
- Support Vector Regression (SVR)
- Least Squares (LS)
- Random Forest (RF)
2.4. SML and DL Modeling
2.5. Metrics
3. Results
4. Discussion
5. Conclusions
Author Contributions
Funding
Institutional Review Board Statement
Informed Consent Statement
Data Availability Statement
Acknowledgments
Conflicts of Interest
Appendix A
Strategy | Component | DL/SML Method | Normal Walk | Slow Walk | Static Situation | Static Situation with CL | Bottom-Top with CL | Left-Right with CL |
---|---|---|---|---|---|---|---|---|
Intras | Fz[N] | ANN | 61.66 (0.976) ± 1.56 (0.001) | 53.45 (0.985) ± 2.49 (0.001) | 13.36 (0.969) ± 0.91 (0.008) | 10.90 (0.991) ± 0.46 (0.002) | 19.14 (0.976) ± 0.47 (0.004) | 25.85 (0.994) ± 0.67 (0.001) |
BLSTM | 63.27 (0.975) ± 2.10 (0.001) | 54.52 (0.984) ± 2.84 (0.002) | 11.64 (0.976) ± 1.10 (0.006) | 12.26 (0.989) ± 0.61 (0.002) | 21.39 (0.970) ± 0.97 (0.004) | 29.04 (0.992) ± 1.14 (0.000) | ||
CNN | 63.38 (0.975) ± 2.27 (0.002) | 52.58 (0.985) ± 2.61 (0.001) | 13.65 (0.968) ± 1.08 (0.008) | 11.47 (0.990) ± 0.60 (0.001) | 19.38 (0.975) ± 0.68 (0.004) | 27.99 (0.993) ± 0.91 (0.000) | ||
SVR | 60.88 (0.977) ± 1.26 (0.001) | 65.57 (0.978) ± 2.88 (0.002) | 14.33 (0.964) ± 0.62 (0.008) | 11.07 (0.991) ± 0.24 (0.002) | 19.47 (0.975) ± 0.58 (0.003) | 26.95 (0.993) ± 0.92 (0.000) | ||
LS | 79.84 (0.961) ± 2.80 (0.002) | 81.67 (0.963) ± 4.31 (0.003) | 26.79 (0.877) ± 2.79 (0.031) | 20.70 (0.968) ± 1.45 (0.007) | 28.89 (0.947) ± 1.32 (0.009) | 49.53 (0.977) ± 2.68 (0.001) | ||
RF | 69.69 (0.971) ± 3.10 (0.002) | 60.15 (0.982) ± 3.63 (0.002) | 11.17 (0.979) ± 1.13 (0.004) | 12.45 (0.988) ± 1.10 (0.003) | 26.78 (0.953) ± 1.71 (0.006) | 33.08 (0.990) ± 1.21 (0.001) | ||
Fy[N] | ANN | 33.40 (0.706) ± 1.68 (0.051) | 30.04 (0.605) ± 1.49 (0.029) | 2.66 (0.761) ± 0.21 (0.015) | 4.00 (0.694) ± 0.53 (0.045) | 8.15 (0.462) ± 0.45 (0.043) | 11.27 (0.635) ± 0.47 (0.031) | |
BLSTM | 35.94 (0.652) ± 2.19 (0.049) | 30.00 (0.575) ± 1.90 (0.027) | 2.14 (0.826) ± 0.21 (0.039) | 3.18 (0.749) ± 0.74 (0.071) | 7.99 (0.445) ± 0.53 (0.049) | 11.89 (0.589) ± 0.66 (0.048) | ||
CNN | 35.36 (0.664) ± 1.81 (0.039) | 31.79 (0.551) ± 2.54 (0.016) | 3.05 (0.715) ± 0.43 (0.058) | 3.92 (0.647) ± 0.38 (0.065) | 8.06 (0.439) ± 0.53 (0.052) | 11.61 (0.612) ± 0.77 (0.048) | ||
SVR | 33.41 (0.705) ± 1.93 (0.051) | 29.29 (0.583) ± 1.36 (0.019) | 6.84 (0.384) ± 0.43 (0.102) | 6.57 (0.422) ± 0.45 (0.098) | 10.42 (0.179) ± 0.38 (0.034) | 13.28 (0.446) ± 0.38 (0.04) | ||
LS | 42.05 (0.581) ± 1.67 (0.047) | 31.19 (0.413) ± 1.64 (0.054) | 4.80 (0.272) ± 0.50 (0.111) | 4.37 (0.319) ± 0.21 (0.126) | 10.12 (−0.018) ± 0.27 (0.052) | 14.89 (0.142) ± 0.43 (0.057) | ||
RF | 33.01 (0.712) ± 1.39 (0.038) | 28.45 (0.598) ± 1.75 (0.015) | 1.97 (0.847) ± 0.26 (0.031) | 4.03 (0.732) ± 0.92 (0.053) | 7.70 (0.474) ± 0.37 (0.039) | 12.44 (0.528) ± 0.80 (0.062) | ||
Fx[N] | ANN | 12.75 (0.884) ± 0.56 (0.012) | 10.42 (0.889) ± 0.55 (0.010) | 2.46 (0.947) ± 0.20 (0.007) | 3.01 (0.938) ± 0.33 (0.014) | 7.03 (0.829) ± 0.53 (0.027) | 8.77 (0.807) ± 0.44 (0.019) | |
BLSTM | 13.27 (0.872) ± 0.61 (0.004) | 9.62 (0.905) ± 0.52 (0.009) | 2.45 (0.948) ± 0.31 (0.013) | 3.14 (0.930) ± 0.40 (0.016) | 7.12 (0.815) ± 0.46 (0.038) | 9.40 (0.775) ± 0.30 (0.014) | ||
CNN | 13.57 (0.867) ± 0.65 (0.010) | 10.79 (0.880) ± 0.94 (0.028) | 3.22 (0.912) ± 0.33 (0.018) | 3.37 (0.924) ± 0.20 (0.011) | 6.91 (0.823) ± 0.53 (0.032) | 9.66 (0.759) ± 0.42 (0.023) | ||
SVR | 14.19 (0.845) ± 0.67 (0.015) | 10.56 (0.885) ± 0.63 (0.007) | 9.32 (0.415) ± 0.78 (0.165) | 5.46 (0.791) ± 0.53 (0.046) | 10.52 (0.641) ± 0.34 (0.046) | 10.30 (0.732) ± 0.33 (0.017) | ||
LS | 18.62 (0.772) ± 1.05 (0.024) | 12.28 (0.844) ± 0.51 (0.020) | 8.43 (0.432) ± 0.60 (0.115) | 6.19 (0.690) ± 0.63 (0.074) | 7.88 (0.776) ± 0.38 (0.034) | 12.76 (0.589) ± 0.55 (0.022) | ||
RF | 13.33 (0.873) ± 0.68 (0.008) | 9.99 (0.895) ± 0.55 (0.008) | 2.59 (0.940) ± 0.14 (0.013) | 2.94 (0.936) ± 0.42 (0.017) | 6.91 (0.832) ± 0.30 (0.032) | 9.67 (0.758) ± 0.39 (0.022) | ||
Inters | Fz[N] | ANN | 92.92 (0.954) ± 20.50 (0.023) | 79.36 (0.972) ± 26.34 (0.011) | 36.64 (0.641) ± 26.40 (0.250) | 42.15(0.698) ± 30.45(0.148) | 45.50 (0.789) ± 19.58 (0.125) | 67.40 (0.984) ± 24.57 (0.007) |
BLSTM | 83.69 (0.956) ± 16.85 (0.023) | 67.14 (0.973) ± 13.50 (0.009) | 24.63 (0.610) ± 9.75 (0.286) | 34.50 (0.582) ± 13.20(0.187) | 53.65 (0.739) ± 21.56 (0.143) | 73.60 (0.973) ± 20.40 (0.015) | ||
CNN | 87.32 (0.955) ± 18.42 (0.025) | 75.38 (0.971) ± 21.62 (0.015) | 24.13 (0.722) ± 11.54 (0.157) | 27.68 (0.726) ± 12.05 (0.183) | 46.82 (0.795) ± 18.35 (0.104) | 69.63 (0.982) ± 30.27 (0.008) | ||
SVR | 60.35 (0.974) ± 21.86 (0.024) | 47.57 (0.986) ± 17.26 (0.012) | 15.64 (0.775) ± 6.09 (0.188) | 15.44 (0.814) ± 7.25 (0.091) | 24.77 (0.909) ± 13.29 (0.050) | 30.91 (0.990) ± 14.00 (0.009) | ||
LS | 84.59 (0.959) ± 25.38 (0.03) | 75.66 (0.969) ± 26.15 (0.021) | 21.33 (0.752) ± 13.07 (0.177) | 20.25 (0.770) ± 8.80 (0.110) | 34.84 (0.838) ± 10.13 (0.102) | 60.98 (0.983) ± 21.00 (0.007) | ||
RF | 52.70 (0.976) ± 27.53 (0.033) | 36.57 (0.989) ± 21.34 (0.016) | 7.97 (0.894) ± 5.75 (0.082) | 15.00 (0.878) ± 25.14 (0.116) | 21.25 (0.921) ± 18.09 (0.052) | 25.51 (0.990) ± 18.41 (0.014) | ||
Fy[N] | ANN | 50.73 (0.402) ± 15.64 (0.236) | 44.85 (0.328) ± 18.58 (0.193) | 16.23 (0.000) ± 17.98 (0.141) | 16.55 (−0.015) ± 18.45 (0.239) | 20.61 (0.126) ± 8.20 (0.338) | 30.71 (−0.146) ± 16.16 (0.291) | |
BLSTM | 44.79 (0.455) ± 17.29 (0.262) | 36.73 (0.430) ± 12.02 (0.165) | 10.97 (−0.048) ± 7.51 (0.223) | 13.22 (−0.047) ± 9.49 (0.225) | 21.10 (−0.024) ± 9.58 (0.316) | 30.67 (−0.126) ± 15.40 (0.278) | ||
CNN | 48.60 (0.437) ± 15.63 (0.158) | 42.79 (0.331) ± 14.87 (0.188) | 10.68 (0.066) ± 6.97 (0.160) | 13.52 (0.159) ± 7.91 (0.220) | 21.32 (0.219) ± 11.42 (0.388) | 31.14 (−0.054) ± 13.29 (0.271) | ||
SVR | 28.27 (0.786) ± 14.79 (0.125) | 20.72 (0.758) ± 7.41 (0.091) | 6.93 (0.026) ± 2.28 (0.220) | 6.63 (−0.128) ± 1.92 (0.240) | 13.00 (0.19) ± 6.90 (0.378) | 14.22 (0.328) ± 8.13 (0.338) | ||
LS | 43.39 (0.527) ± 15.35 (0.151) | 31.49 (0.360) ± 10.63 (0.172) | 7.53 (0.061) ± 5.23 (0.251) | 6.70 (0.048) ± 3.15 (0.368) | 13.29 (−0.281) ± 6.68 (0.293) | 6.89 (−0.068) ± 4.62 (0.337) | ||
RF | 22.31 (0.876) ± 14.36 (0.088) | 16.35 (0.850) ± 7.49 (0.082) | 1.35 (0.571) ± 1.27 (0.465) | 2.81 (0.660) ± 4.32 (0.281) | 7.31 (0.662) ± 8.11 (0.369) | 9.60 (0.652) ± 7.50 (0.395) | ||
Fx[N] | ANN | 19.72 (0.716) ± 5.58 (0.098) | 14.74 (0.753) ± 4.64 (0.131) | 10.22 (0.074) ± 4.39 (0.380) | 10.58 (0.227) ± 7.27 (0.202) | 16.46 (0.230) ± 6.58 (0.315) | 18.28 (0.527) ± 7.73 (0.236) | |
BLSTM | 17.42 (0.765) ± 4.63 (0.094) | 12.58 (0.819) ± 2.47 (0.075) | 11.07 (0.140) ± 3.82 (0.165) | 7.42 (0.158) ± 2.61 (0.201) | 14.58 (0.192) ± 4.63 (0.282) | 15.67 (0.514) ± 5.83 (0.268) | ||
CNN | 18.79 (0.717) ± 5.06 (0.090) | 13.56 (0.792) ± 3.43 (0.078) | 9.53 (0.221) ± 4.11 (0.241) | 7.09 (0.015) ± 2.25 (0.238) | 13.31 (0.235) ± 6.06 (0.260) | 16.38 (0.523) ± 6.24 (0.261) | ||
SVR | 13.06 (0.865) ± 4.06 (0.069) | 10.48 (0.874) ± 1.79 (0.043) | 10.72 (0.144) ± 5.79 (0.313) | 5.29 (0.312) ± 1.88 (0.304) | 9.56 (0.280) ± 2.74 (0.218) | 9.90 (0.646) ± 2.06 (0.198) | ||
LS | 19.01 (0.766) ± 3.74 (0.083) | 12.50 (0.817) ± 1.72 (0.064) | 9.03 (0.048) ± 5.75 (0.234) | 8.00 (0.193) ± 3.67 (0.186) | 9.30 (0.368) ± 4.90 (0.136) | 14.81 (0.539) ± 6.25 (0.208) | ||
RF | 9.29 (0.924) ± 4.88 (0.071) | 6.62 (0.945) ± 3.48 (0.041) | 1.65 (0.602) ± 1.59 (0.277) | 1.33 (0.696) ± 0.99 (0.153) | 4.39 (0.651) ± 2.50 (0.184) | 6.43 (0.796) ± 4.12 (0.252) |
Strategy | Component | DL/SML Method | Normal Walk | Slow Walk | Static Situation | Static Situation with CL | Bottom-Top with CL | Left-Right with CL |
---|---|---|---|---|---|---|---|---|
Intras | Fz[N] | ANN | 56.60 (0.981) ± 2.63 (0.001) | 50.48 (0.985) ± 2.98 (0.001) | 12.44 (0.978) ± 0.62 (0.005) | 12.56 (0.977) ± 0.65 (0.003) | 19.84 (0.968) ± 1.09 (0.003) | 27.10 (0.992) ± 0.78 (0.001) |
BLSTM | 56.70 (0.981) ± 3.20 (0.002) | 51.18 (0.985) ± 4.16 (0.002) | 11.42 (0.981) ± 0.69 (0.005) | 13.58 (0.974) ± 1.21 (0.003) | 22.04 (0.961) ± 1.44 (0.005) | 30.97 (0.99) ± 3.36 (0.002) | ||
CNN | 57.94 (0.98) ± 3.56 (0.002) | 51.86 (0.984) ± 2.81 (0.002) | 13.23 (0.976) ± 0.98 (0.004) | 13.17 (0.975) ± 0.63 (0.002) | 21.77 (0.962) ± 1.83 (0.006) | 28.77 (0.991) ± 0.59 (0.001) | ||
SVR | 61.26 (0.979) ± 1.41 (0.001) | 49.31 (0.986) ± 1.67 (0.001) | 13.23 (0.974) ± 0.53 (0.004) | 13.85 (0.973) ± 0.52 (0.004) | 23.35 (0.957) ± 1.59 (0.005) | 29.45 (0.991) ± 0.97 (0.001) | ||
LS | 74.89 (0.967) ± 2.18 (0.002) | 71.60 (0.971) ± 2.12 (0.001) | 27.60 (0.909) ± 1.22 (0.026) | 26.70 (0.916) ± 1.68 (0.019) | 36.72 (0.892) ± 2.24 (0.016) | 44.32 (0.980) ± 1.72 (0.002) | ||
RF | 71.16 (0.971) ± 3.62 (0.002) | 59.08 (0.980) ± 2.79 (0.002) | 11.29 (0.981) ± 0.73 (0.003) | 14.89 (0.969) ± 0.89 (0.004) | 27.04 (0.942) ± 2.05 (0.008) | 37.09 (0.986) ± 1.13 (0.001) | ||
Fy[N] | ANN | 33.55 (0.685) ± 2.01 (0.040) | 24.45 (0.715) ± 1.17 (0.024) | 3.67 (0.569) ± 0.42 (0.091) | 3.51 (0.703) ± 0.40 (0.116) | 8.28 (0.459) ± 0.39 (0.049) | 11.33 (0.623) ± 0.57 (0.031) | |
BLSTM | 37.04 (0.599) ± 1.95 (0.049) | 24.55 (0.707) ± 1.10 (0.025) | 3.03 (0.678) ± 0.74 (0.130) | 3.07 (0.761) ± 0.51 (0.097) | 8.41 (0.432) ± 0.45 (0.039) | 11.44 (0.598) ± 0.47 (0.056) | ||
CNN | 36.39 (0.615) ± 1.73 (0.044) | 25.83 (0.672) ± 2.17 (0.041) | 4.16 (0.460) ± 0.60 (0.140) | 3.88 (0.635) ± 0.30 (0.086) | 8.46 (0.414) ± 0.46 (0.035) | 11.96 (0.549) ± 0.73 (0.065) | ||
SVR | 34.51 (0.659) ± 1.54 (0.032) | 22.12 (0.759) ± 0.47 (0.011) | 7.99 (0.092) ± 0.25 (0.132) | 8.01 (0.093) ± 0.43 (0.118) | 9.31 (0.292) ± 0.37 (0.036) | 13.47 (0.390) ± 0.69 (0.058) | ||
LS | 42.57 (0.369) ± 1.31 (0.047) | 27.53 (0.563) ± 1.16 (0.009) | 6.55 (−0.109) ± 0.59 (0.146) | 7.50 (−0.442) ± 0.41 (0.054) | 9.96 (0.119) ± 0.41 (0.057) | 14.83 (0.079) ± 0.55 (0.074) | ||
RF | 39.03 (0.543) ± 1.87 (0.046) | 22.82 (0.734) ± 0.90 (0.017) | 3.84 (0.563) ± 0.59 (0.110) | 2.92 (0.768) ± 0.39 (0.085) | 8.55 (0.359) ± 0.38 (0.053) | 13.13 (0.434) ± 0.75 (0.052) | ||
Fx[N] | ANN | 17.46 (0.621) ± 1.26 (0.033) | 10.47 (0.836) ± 0.57 (0.018) | 3.41 (0.929) ± 0.25 (0.017) | 3.58 (0.927) ± 0.53 (0.022) | 6.64 (0.858) ± 0.46 (0.022) | 9.28 (0.842) ± 0.52 (0.015) | |
BLSTM | 16.89 (0.628) ± 1.03 (0.023) | 10.12 (0.846) ± 0.36 (0.008) | 3.27 (0.936) ± 0.24 (0.010) | 3.47 (0.935) ± 0.34 (0.011) | 7.00 (0.840) ± 0.37 (0.031) | 9.88 (0.818) ± 0.38 (0.016) | ||
CNN | 17.56 (0.594) ± 0.90 (0.025) | 10.82 (0.821) ± 0.36 (0.018) | 3.95 (0.914) ± 0.53 (0.017) | 4.29 (0.895) ± 0.27 (0.015) | 6.97 (0.839) ± 0.57 (0.030) | 10.01 (0.818) ± 0.54 (0.011) | ||
SVR | 16.29 (0.642) ± 0.857 (0.017) | 10.24 (0.839) ± 0.148 (0.017) | 9.08 (0.549) ± 0.47 (0.143) | 8.29 (0.665) ± 0.62 (0.046) | 10.10 (0.659) ± 0.32 (0.052) | 10.95 (0.769) ± 0.34 (0.021) | ||
LS | 18.09 (0.613) ± 0.76 (0.018) | 11.62 (0.791) ± 0.20 (0.019) | 6.08 (0.779) ± 0.40 (0.051) | 7.00 (0.721) ± 0.59 (0.056) | 8.60 (0.742) ± 0.42 (0.042) | 13.29 (0.661) ± 0.59 (0.029) | ||
RF | 17.42 (0.584) ± 0.94 (0.025) | 9.87 (0.849) ± 0.27 (0.017) | 2.73 (0.950) ± 0.24 (0.010) | 3.46 (0.931) ± 0.35 (0.011) | 6.77 (0.850) ± 0.37 (0.027) | 10.40 (0.797) ± 0.46 (0.015) | ||
Inters | Fz[N] | ANN | 99.22 (0.945) ± 33.67 (0.023) | 90.54 (0.958) ± 25.95 (0.014) | 44.55 (0.582) ± 26.22 (0.203) | 49.28 (0.563) ± 23.17 (0.189) | 47.17 (0.765) ± 17.17 (0.122) | 56.15 (0.976) ± 9.94 (0.012) |
BLSTM | 81.82 (0.958) ± 23.63 (0.024) | 76.53 (0.967) ± 17.79 (0.015) | 46.49 (0.610) ± 26.05 (0.162) | 41.30 (0.561) ± 22.93 (0.164) | 54.41 (0.781) ± 26.75 (0.114) | 61.89 (0.977) ± 20.62 (0.011) | ||
CNN | 90.18 (0.954) ± 26.81 (0.019) | 84.33 (0.959) ± 19.68 (0.016) | 49.60 (0.468) ± 22.45 (0.270) | 50.34 (0.547) ± 25.04 (0.170) | 49.48 (0.808) ± 24.70 (0.078) | 57.22 (0.982) ± 14.71 (0.008) | ||
SVR | 65.58 (0.969) ± 26.53 (0.027) | 59.44 (0.978) ± 30.01 (0.019) | 32.96 (0.746) ± 34.13 (0.194) | 28.35 (0.742) ± 28.13 (0.147) | 30.62 (0.887) ± 23.70 (0.061) | 32.60 (0.989) ± 18.06 (0.011) | ||
LS | 86.04 (0.960) ± 23.95 (0.027) | 85.54 (0.964) ± 24.22 (0.018) | 37.20 (0.703) ± 19.36 (0.202) | 38.65 (0.685) ± 23.69 (0.178) | 53.48 (0.846) ± 18.00 (0.090) | 59.58 (0.982) ± 19.79 (0.009) | ||
RF | 51.17 (0.976) ± 28.68 (0.033) | 38.99 (0.987) ± 22.74 (0.017) | 12.59 (0.878) ± 18.41 (0.121) | 9.20 (0.889) ± 6.74 (0.082) | 19.51 (0.921) ± 12.73 (0.051) | 28.01 (0.989) ± 21.20 (0.015) | ||
Fy[N] | ANN | 49.40 (0.472) ± 18.14 (0.218) | 36.12 (0.542) ± 8.42 (0.110) | 17.23 (−0.055) ± 12.22 (0.247) | 14.53 (−0.096) ± 12.53 (0.341) | 19.90 (−0.053) ± 10.55 (0.200) | 25.67 (−0.191) ± 9.55 (0.146) | |
BLSTM | 41.37 (0.538) ± 9.71 (0.107) | 32.98 (0.559) ± 8.53 (0.121) | 14.12 (−0.055) ± 12.56 (0.209) | 13.51 (−0.105) ± 12.73 (0.295) | 16.11 (−0.07) ± 8.63 (0.180) | 25.57 (−0.060) ± 13.27 (0.204) | ||
CNN | 46.25 (0.455) ± 15.78 (0.254) | 38.01 (0.503) ± 12.49 (0.250) | 14.86 (0.070) ± 9.48 (0.291) | 13.81 (−0.035) ± 10.60 (0.389) | 14.93 (0.064) ± 7.54 (0.256) | 21.50 (0.030) ± 9.17 (0.275) | ||
SVR | 27.65 (0.817) ± 12.60 (0.075) | 19.94 (0.810) ± 7.11 (0.089) | 8.98 (−0.216) ± 6.38 (0.309) | 12.09 (−0.095) ± 7.97 (0.347) | 9.35 (0.143) ± 3.84 (0.263) | 12.77 (0.452) ± 6.27 (0.258) | ||
LS | 42.55 (0.483) ± 13.72 (0.183) | 29.24 (0.600) ± 8.27 (0.109) | 9.50 (−0.234) ± 7.59 (0.143) | 9.62 (−0.346) ± 8.83 (0.310) | 11.717 (−0.148) ± 4.55 (0.213) | 17.27 (−0.243) ± 5.28 (0.224) | ||
RF | 23.32 (0.874) ± 11.10 (0.055) | 15.31 (0.893) ± 6.38 (0.053) | 3.56 (0.585) ± 5.77 (0.267) | 1.94 (0.544) ± 1.48 (0.357) | 6.07 (0.580) ± 3.70 (0.345) | 10.02 (0.720) ± 8.45 (0.269) | ||
Fx[N] | ANN | 23.34 (0.406) ± 8.16 (0.227) | 18.33 (0.514) ± 7.78 (0.228) | 12.86 (0.124) ± 8.71 (0.143) | 13.39 (0.032) ± 7.34 (0.220) | 10.85 (0.277) ± 4.73 (0.202) | 17.67 (0.350) ± 2.73 (0.267) | |
BLSTM | 18.63 (0.513) ± 5.38 (0.212) | 12.84 (0.681) ± 2.34 (0.098) | 8.35 (0.13) ± 4.84 (0.254) | 9.79 (0.031) ± 5.15 (0.247) | 12.26 (0.265) ± 4.08 (0.122) | 15.74 (0.369) ± 4.14 (0.322) | ||
CNN | 20.05 (0.473) ± 5.23 (0.198) | 14.91 (0.606) ± 3.25 (0.149) | 9.11 (−0.045) ± 5.10 (0.203) | 11.73 (0.116) ± 6.29 (0.242) | 11.23 (0.294) ± 4.46 (0.109) | 16.26 (0.414) ± 4.15 (0.260) | ||
SVR | 14.63 (0.672) ± 4.95 (0.223) | 10.34 (0.772) ± 1.57 (0.117) | 10.23 (0.151) ± 4.30 (0.235) | 8.39 (0.239) ± 4.02 (0.264) | 9.64 (0.323) ± 2.791 (0.133) | 10.50 (0.524) ± 2.37 (0.262) | ||
LS | 17.22 (0.578) ± 4.85 (0.146) | 13.27 (0.703) ± 3.33 (0.097) | 7.50 (0.114) ± 3.32 (0.167) | 8.20 (0.188) ± 4.22 (0.157) | 9.16 (0.433) ± 4.44 (0.136) | 14.88 (0.508) ± 4.29 (0.195) | ||
RF | 7.01 (0.803) ± 3.08 (0.149) | 4.43 (0.681) ± 3.16 (0.233) | 2.29 (0.562) ± 2.41 (0.269) | 1.78 (0.538) ± 1.44 (0.340) | 7.64 (0.888) ± 4.82 (0.114) | 12.09 (0.760) ± 6.90 (0.262) |
Appendix B
References
- Logar, G.; Munih, M. Estimation of Joint Forces and Moments for the In-Run and Take-Off in Ski Jumping Based on Measurements with Wearable Inertial Sensors. Sensors 2015, 15, 11258–11276. [Google Scholar] [CrossRef]
- Hori, N.; Newton, R.U.; Kawamori, N.; McGuigan, M.R.; Kraemer, W.J.; Nosaka, K. Reliability of Performance Measurements Derived from Ground Reaction Force Data during Countermovement Jump and the Influence of Sampling Frequency. J. Strength Cond. Res. 2009, 23, 874–882. [Google Scholar] [CrossRef]
- Ericksen, H.M.; Gribble, P.A.; Pfile, K.R.; Pietrosimone, B. Different Modes of Feedback and Peak Vertical Ground Reaction Force during Jump Landing: A Systematic Review. J. Athl. Train. 2013, 48, 685–695. [Google Scholar] [CrossRef] [PubMed]
- Fregly, B.J.; Reinbolt, J.A.; Rooney, K.L.; Mitchell, K.H.; Chmielewski, T.L. Design of Patient-Specific Gait Modifications for Knee Osteoarthritis Rehabilitation. IEEE Trans. Biomed. Eng. 2007, 54, 1687–1695. [Google Scholar] [CrossRef]
- Houck, J.; Kneiss, J.A.; Bukata, S.V.; Puzas, J.E. Analysis of Vertical Ground Reaction Force Variables during a Sit to Stand Task in Participants Recovering from a Hip Fracture. Clin. Biomech. 2011, 26, 470–476. [Google Scholar] [CrossRef]
- Shin, K.Y.; Rim, Y.H.; Kim, Y.S.; Kim, H.S.; Han, J.S.; Park, C.H.; Lee, K.S.; Mun, J.H. A Joint Normalcy Index to Evaluate Patients with Gait Pathologies in the Functional Aspects of Joint Mobility. J. Mech. Sci. Technol. 2010, 24, 1901–1909. [Google Scholar] [CrossRef]
- Sim, T.; Kwon, H.; Oh, S.E.; Joo, S.; Choi, A.; Heo, H.; Kim, K.-S.; Mun, J.H. Predicting Complete Ground Reaction Forces and Moments during GAIT with Insole Plantar Pressure Information Using a Wavelet Neural Network. J. Biomech. Eng. 2015, 137, 091001. [Google Scholar] [CrossRef]
- Parry, R.; Lalo, E.; Roussel, J.; Jabloun, M.; Riff, J.; Welter, M.; Buttelli, O. Caractérisation Du Freezing de La Marche En Situation Réelle. Neurophysiol. Clin. 2015, 45, 392. [Google Scholar] [CrossRef]
- Parry, R.; Sellam, N.; Lalo, E.; Welter, M.; Buttelli, O. Pattern Électromyographique de La Marche Parkinsonienne En Condition de Vie Réelle. Neurophysiol. Clin. 2016, 46, 273–274. [Google Scholar] [CrossRef]
- Odebiyi, D.O.; Okafor, U.A.C. Musculoskeletal Disorders, Workplace Ergonomics and Injury Prevention. In IntechOpen eBooks; IntechOpen: London, UK, 2023. [Google Scholar]
- Yazji, M.; Raison, M.; Aubin, C.-É.; Labelle, H.; Detrembleur, C.; Mahaudens, P.; Mousny, M. Are the Mediolateral Joint Forces in the Lower Limbs Different between Scoliotic and Healthy Subjects during Gait? Scoliosis 2015, 10, S3. [Google Scholar] [CrossRef] [PubMed]
- Bezzini, R.; Crosato, L.; Losè, M.T.; Avizzano, C.A.; Bergamasco, M.; Filippeschi, A. Closed-Chain Inverse Dynamics for the Biomechanical Analysis of Manual Material Handling Tasks through a Deep Learning Assisted Wearable Sensor Network. Sensors 2023, 23, 5885. [Google Scholar] [CrossRef]
- Matijevich, E.S.; Völgyesi, P.; Zelik, K.E. A Promising Wearable Solution for the Practical and Accurate Monitoring of Low Back Loading in Manual Material Handling. Sensors 2021, 21, 340. [Google Scholar] [CrossRef]
- Larsen, F.G.; Svenningsen, F.P.; Andersen, M.S.; De Zee, M.; Skals, S. Estimation of Spinal Loading during Manual Materials Handling Using Inertial Motion Capture. Ann. Biomed. Eng. 2019, 48, 805–821. [Google Scholar] [CrossRef] [PubMed]
- Muller, A.; Pontonnier, C.; Robert-Lachaîne, X.; Dumont, G.; Plamondon, A. Motion-Based Prediction of External Forces and Moments and Back Loading during Manual Material Handling Tasks. Appl. Ergon. 2020, 82, 102935. [Google Scholar] [CrossRef]
- Corbeil, P.; Plamondon, A.; Handrigan, G.; Vallée-Marcotte, J.; Laurendeau, S.; Have, J.T.; Manzerolle, N. Biomechanical Analysis of Manual Material Handling Movement in Healthy Weight and Obese Workers. Appl. Ergon. 2019, 74, 124–133. [Google Scholar] [CrossRef]
- Gagnon, D.; Plamondon, A.; Larivière, C. A Comparison of Lumbar Spine and Muscle Loading between Male and Female Workers during Box Transfers. J. Biomech. 2018, 81, 76–85. [Google Scholar] [CrossRef]
- Savelberg, H.H.C.M.; Lange, A. Assessment of the Horizontal, Fore-Aft Component of the Ground Reaction Force from Insole Pressure Patterns by Using Artificial Neural Networks. Clin. Biomech. 1999, 14, 585–592. [Google Scholar] [CrossRef] [PubMed]
- Fong, D.T.-P.; Chan, Y.M.; Hong, Y.; Yung, P.S.; Fung, K.P.; Chan, K. Estimating the Complete Ground Reaction Forces with Pressure Insoles in Walking. J. Biomech. 2008, 41, 2597–2601. [Google Scholar] [CrossRef]
- Rouhani, H.; Favre, J.; Crevoisier, X.; Aminian, K. Ambulatory Assessment of 3D Ground Reaction Force Using Plantar Pressure Distribution. Gait Posture 2010, 32, 311–316. [Google Scholar] [CrossRef] [PubMed]
- Jacobs, D.A.; Ferris, D.P. Estimation of Ground Reaction Forces and Ankle Moment with Multiple, Low-Cost Sensors. J. Neuroeng. Rehabil. 2015, 12, 90. [Google Scholar] [CrossRef]
- Joo, S.; Oh, S.E.; Mun, J.H. Improving the Ground Reaction Force Prediction Accuracy Using One-Axis Plantar Pressure: Expansion of Input Variable for Neural Network. J. Biomech. 2016, 49, 3153–3161. [Google Scholar] [CrossRef]
- Honert, E.C.; Hoitz, F.; Blades, S.; Nigg, S.; Nigg, B.M. Estimating Running Ground Reaction Forces from Plantar Pressure during Graded Running. Sensors 2022, 22, 3338. [Google Scholar] [CrossRef] [PubMed]
- Kammoun, A.; Ravier, P.; Buttelli, O. Impact of PCA Pre-Normalization Methods on Ground Reaction Force Estimation Accuracy. Sensors 2024, 24, 1137. [Google Scholar] [CrossRef]
- Taye, M.M. Understanding of Machine Learning with Deep Learning: Architectures, Workflow, Applications and Future Directions. Computers 2023, 12, 91. [Google Scholar] [CrossRef]
- Wu, D.; Jennings, C.; Terpenny, J.; Gao, R.X.; Kumara, S. A Comparative Study on Machine Learning Algorithms for Smart Manufacturing: Tool Wear Prediction Using Random Forests. J. Manuf. Sci. Eng. 2017, 139, 071018. [Google Scholar] [CrossRef]
- Nawar, S.; Mouazen, A.M. Comparison between Random Forests, Artificial Neural Networks and Gradient Boosted Machines Methods of On-Line Vis-NIR Spectroscopy Measurements of Soil Total Nitrogen and Total Carbon. Sensors 2017, 17, 2428. [Google Scholar] [CrossRef]
- Belina, Y.; Kebede, A. Comparative Study of Artificial Neural Network (ANN) and Support Vector Regression (SVR) in Rainfall-Runoff Modeling of Awash Belo Watershed, Awash River Basin, Ethiopia. Preprint 2023. [Google Scholar] [CrossRef]
- World Medical Association. World Medical Association Declaration of Helsinki. JAMA 2013, 310, 2191. [Google Scholar] [CrossRef]
- Alcantara, R.S.; Edwards, W.B.; Millet, G.Y.; Grabowski, A.M. Predicting Continuous Ground Reaction Forces from Accelerometers during Uphill and Downhill Running: A Recurrent Neural Network Solution. PeerJ 2022, 10, e12752. [Google Scholar] [CrossRef] [PubMed]
- Haidar, A.; Verma, B. Monthly Rainfall Forecasting Using One-Dimensional Deep Convolutional Neural Network. IEEE Access 2018, 6, 69053–69063. [Google Scholar] [CrossRef]
- Pangarkar, D.J.; Sharma, R.; Sharma, A.; Sharma, M. Assessment of the Different Machine Learning Models for Prediction of Cluster Bean (Cyamopsis tetragonoloba L. Taub.) Yield. Adv. Res. 2020, 21, 98–105. [Google Scholar] [CrossRef]
ANN | CNN | BLSTM | SVR | RF |
---|---|---|---|---|
2 hidden layers of (256, 128) neurons, activation function: leaky relu, normalization method: BW, batch size: 4, optimizer: Adamax, learning rate: 0.01 | 1 convolution layer: Conv1D (number of filters = 8, kernel size = 4), 1 MaxPool1d layer (kernel size = 2), an ANN with 3 hidden layers of (96, 60, 25) neurons | 2 BLSTM layers with (400, 200) cells, an ANN with 3 hidden layers of (400, 300, 150) neurons, an input sequence size equal to 20 | RBF kernel model with = 20, C = 100, and = 0.1 | T = 1000 |
Activity | Right Foot | Left Foot |
---|---|---|
Total Dataset: 160,183 | Total Dataset: 164,526 | |
Normal walk | 11,100 (101) | 12,719 (111) |
Slow walk | 15,008 (112) | 17,732 (128) |
Static situation | 27,032 | 27,032 |
Static situation with CL | 24,197 | 24,197 |
Bottom-top with CL | 41,777 (86) | 41,777 (86) |
Left-right with CL | 41,069 (72) | 41,069 (72) |
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content. |
© 2024 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/).
Share and Cite
Kammoun, A.; Ravier, P.; Buttelli, O. Comparison of the Accuracy of Ground Reaction Force Component Estimation between Supervised Machine Learning and Deep Learning Methods Using Pressure Insoles. Sensors 2024, 24, 5318. https://doi.org/10.3390/s24165318
Kammoun A, Ravier P, Buttelli O. Comparison of the Accuracy of Ground Reaction Force Component Estimation between Supervised Machine Learning and Deep Learning Methods Using Pressure Insoles. Sensors. 2024; 24(16):5318. https://doi.org/10.3390/s24165318
Chicago/Turabian StyleKammoun, Amal, Philippe Ravier, and Olivier Buttelli. 2024. "Comparison of the Accuracy of Ground Reaction Force Component Estimation between Supervised Machine Learning and Deep Learning Methods Using Pressure Insoles" Sensors 24, no. 16: 5318. https://doi.org/10.3390/s24165318