Figure 1.
Two subjects during the cognitive load session. Left: reporting the results of the math calculation. Right: during the audio listening task.
Figure 1.
Two subjects during the cognitive load session. Left: reporting the results of the math calculation. Right: during the audio listening task.
Figure 2.
Controlled experiment. Top left: the plant of the laboratory, where the path chosen for the walking activities is depicted. A red rectangle identifies the collision avoidance zone in the two images of the first raw. In this zone, two obstacles are moved by one of the experimenters and the two subjects have to avoid the collision (figure bottom right). During the rest of the path, subjects walk with their own natural pace.
Figure 2.
Controlled experiment. Top left: the plant of the laboratory, where the path chosen for the walking activities is depicted. A red rectangle identifies the collision avoidance zone in the two images of the first raw. In this zone, two obstacles are moved by one of the experimenters and the two subjects have to avoid the collision (figure bottom right). During the rest of the path, subjects walk with their own natural pace.
Figure 3.
Wearable devices adopted.
Figure 3.
Wearable devices adopted.
Figure 4.
Boxplots of the peak rate in the different tasks. The left boxplot in green (A) refers to the young adults while the right one in blue (B) refers to the older adults. For each box, the outliers are marked using the ’+’ symbol.
Figure 4.
Boxplots of the peak rate in the different tasks. The left boxplot in green (A) refers to the young adults while the right one in blue (B) refers to the older adults. For each box, the outliers are marked using the ’+’ symbol.
Table 1.
Number of instances for each task in cognitive load session (first 5 columns) and in walking session (last 8 columns). For each subject group (Young Adults and Older Adults), the first row is related to signals collected using the Shimmer3 GSR+ unit (PPG and GSR) while the second row refers to signals acquired using the Shimmer3 EMG (gastrocnemius muscle EMG and tibial muscle EMG). In the third row is reported the cardinality involved during the classification tasks. In particular, for the reading, comprehension and free walk tasks are reported the cardinality resulting from data augmentation. The analyzed tasks have been referred to according to the following coding: BC = baseline task collected during the cognitive load session, R = reading task, C = calculation task, AL = audio listening task, BW = baseline task collected during the walking session, F1 = metronome forced speed task (70 bpm), F2 = metronome forced speed task (85 bpm), F3 = metronome forced speed task (100 bpm), FW = pure free walk task, WO = free walk in the collision avoidance task, Obs = obstacle crossing, WO + Obs = single signal for the whole task of collision avoidance (free walk and obstacle crossing).
Table 1.
Number of instances for each task in cognitive load session (first 5 columns) and in walking session (last 8 columns). For each subject group (Young Adults and Older Adults), the first row is related to signals collected using the Shimmer3 GSR+ unit (PPG and GSR) while the second row refers to signals acquired using the Shimmer3 EMG (gastrocnemius muscle EMG and tibial muscle EMG). In the third row is reported the cardinality involved during the classification tasks. In particular, for the reading, comprehension and free walk tasks are reported the cardinality resulting from data augmentation. The analyzed tasks have been referred to according to the following coding: BC = baseline task collected during the cognitive load session, R = reading task, C = calculation task, AL = audio listening task, BW = baseline task collected during the walking session, F1 = metronome forced speed task (70 bpm), F2 = metronome forced speed task (85 bpm), F3 = metronome forced speed task (100 bpm), FW = pure free walk task, WO = free walk in the collision avoidance task, Obs = obstacle crossing, WO + Obs = single signal for the whole task of collision avoidance (free walk and obstacle crossing).
| | Cognitive Load Session | Walking Session |
---|
| | BC | R | C | MC | AL | BW | F1 | F2 | F3 | FW | WO | Obs | WO + Obs |
---|
Young Adults | PPG, GSR | 46 | 32 | 32 | 96 | 96 | 109 | 46 | 46 | 46 | - | - | - | 46 |
Gastrocn. EMG, Tibial EMG | - | - | - | - | - | 55 | 23 | 23 | 23 | - | - | - | 23 |
PPG, GSR (augmented) | 46 | 96 | 64 | 96 | 96 | 109 | 46 | 46 | 46 | - | - | - | 46 |
Older Adlults | PPG, GSR | 60 | 40 | 40 | 120 | 120 | 129 | 57 | 57 | 57 | 57 | 160 | 104 | - |
Gastrocn. EMG, Tibial EMG | - | - | - | - | - | 65 | 28 | 27 | 28 | 28 | 77 | 50 | - |
PPG, GSR (augmented) | 60 | 120 | 80 | 120 | 120 | 129 | 57 | 57 | 57 | 114 | 160 | 104 | - |
Table 2.
Performance of the binary classifiers in discriminating young adults (Yng) from older adults (Old) in the different tasks analyzed, varying the feature set and adopting a LOSO validation strategy. Three performance metrics are reported: accuracy (Acc.), F1-score () and weighted F1-score (W-). In each table, the W- values in bold represent the best performances achieved for each feature set considered, while in red is highlighted the best accuracy at all. The analyzed tasks have been referred according to the following coding: R = reading task, C = comprehension task, MC = math calculation task, AL = audio listening task.
Table 2.
Performance of the binary classifiers in discriminating young adults (Yng) from older adults (Old) in the different tasks analyzed, varying the feature set and adopting a LOSO validation strategy. Three performance metrics are reported: accuracy (Acc.), F1-score () and weighted F1-score (W-). In each table, the W- values in bold represent the best performances achieved for each feature set considered, while in red is highlighted the best accuracy at all. The analyzed tasks have been referred according to the following coding: R = reading task, C = comprehension task, MC = math calculation task, AL = audio listening task.
| | R Task | C Task | MC Task | AL Task |
---|
| Classifier | | Yng | Old | | | Yng | Old | | | Yng | Old | | | Yng | Old | |
Acc. | | | W- | Acc. | | | W- | Acc. | | | W- | Acc. | | | W- |
PPG | SVM Linear | 63% | 0.53 | 0.69 | 62% | 69% | 0.55 | 0.76 | 67% | 74% | 0.68 | 0.78 | 73% | 60% | 0.44 | 0.69 | 58% |
SVM Cubic | 64% | 0.59 | 0.68 | 64% | 62% | 0.60 | 0.64 | 62% | 70% | 0.68 | 0.72 | 70% | 64% | 0.62 | 0.65 | 64% |
SVM Gauss | 66% | 0.61 | 0.70 | 66% | 65% | 0.41 | 0.75 | 60% | 71% | 0.65 | 0.75 | 71% | 56% | 0.39 | 0.66 | 54% |
Cart | 56% | 0.54 | 0.59 | 57% | 59% | 0.56 | 0.62 | 59% | 73% | 0.68 | 0.76 | 73% | 50% | 0.46 | 0.52 | 50% |
Random | 50% | 0.47 | 0.52 | 50% | 50% | 0.44 | 0.55 | 50% | 50% | 0.47 | 0.52 | 50% | 50% | 0.47 | 0.52 | 50% |
GSR | SVM Linear | 63% | 0.54 | 0.70 | 63% | 66% | 0.63 | 0.69 | 66% | 64% | 0.59 | 0.68 | 64% | 67% | 0.60 | 0.72 | 67% |
SVM Cubic | 58% | 0.53 | 0.62 | 58% | 58% | 0.53 | 0.61 | 58% | 66% | 0.61 | 0.70 | 66% | 58% | 0.51 | 0.63 | 58% |
SVM Gauss | 62% | 0.54 | 0.68 | 62% | 65% | 0.59 | 0.70 | 65% | 62% | 0.55 | 0.67 | 62% | 59% | 0.49 | 0.66 | 58% |
Cart | 59% | 0.52 | 0.64 | 59% | 63% | 0.57 | 0.67 | 62% | 54% | 0.49 | 0.59 | 54% | 57% | 0.54 | 0.61 | 58% |
Random | 50% | 0.47 | 0.52 | 50% | 50% | 0.44 | 0.55 | 50% | 50% | 0.47 | 0.52 | 50% | 50% | 0.47 | 0.52 | 50% |
PPG and GSR | SVM Linear | 75% | 0.71 | 0.77 | 74% | 69% | 0.67 | 0.72 | 70% | 78% | 0.75 | 0.81 | 78% | 65% | 0.59 | 0.69 | 65% |
SVM Cubic | 64% | 0.62 | 0.66 | 64% | 61% | 0.56 | 0.65 | 61% | 69% | 0.64 | 0.72 | 68% | 64% | 0.58 | 0.69 | 64% |
SVM Gauss | 72% | 0.67 | 0.67 | 72% | 61% | 0.56 | 0.65 | 61% | 71% | 0.68 | 0.74 | 71% | 65% | 0.59 | 0.69 | 65% |
Cart | 65% | 0.60 | 0.69 | 65% | 58% | 0.52 | 0.63 | 58% | 67% | 0.63 | 0.70 | 67% | 62% | 0.55 | 0.66 | 61% |
Random | 50% | 0.47 | 0.52 | 50% | 50% | 0.44 | 0.55 | 50% | 50% | 0.47 | 0.52 | 50% | 50% | 0.47 | 0.52 | 50% |
Table 3.
Confusion matrix of SVM-Linear for the multi-class recognition task. Six classes are considered, one for each couple of the cognitive load task: math calculation MC, reading R and audio listening AL, and subject age: young adult (Yng), and older adults (Old). The values in bold are the main diagonal elements and represent the cases where the classes predicted by the classifier and true classes agree.
Table 3.
Confusion matrix of SVM-Linear for the multi-class recognition task. Six classes are considered, one for each couple of the cognitive load task: math calculation MC, reading R and audio listening AL, and subject age: young adult (Yng), and older adults (Old). The values in bold are the main diagonal elements and represent the cases where the classes predicted by the classifier and true classes agree.
| | Predicted Class |
---|
| | MC_Yng | R_Yng | AL_Yng | MC_Old | R_Old | AL_Old |
True class | MC_Yng | 56% | 7% | 3% | 19% | 10% | 4% |
R_Yng | 4% | 46% | 2% | 5% | 26% | 17% |
AL_Yng | 0% | 6% | 50% | 0% | 4% | 40% |
MC_Old | 14% | 7% | 1% | 71% | 1% | 7% |
R_Old | 2% | 15% | 2% | 3% | 76% | 3% |
AL_Old | 2% | 12% | 18% | 2% | 2% | 66% |
Table 4.
Performance of the binary classifiers in discriminating the instances collected during the math calculation tasks MC from the instances collected during the audio listening task AL, varying the feature set and adopting a LOSO validation strategy. Three performance metrics are reported: accuracy (Acc.), F1-score () and weighted F1-score (W-). In each population group, the W- values in bold represent the best performances achieved for each feature set considered, while in red is highlighted the best accuracy at all.
Table 4.
Performance of the binary classifiers in discriminating the instances collected during the math calculation tasks MC from the instances collected during the audio listening task AL, varying the feature set and adopting a LOSO validation strategy. Three performance metrics are reported: accuracy (Acc.), F1-score () and weighted F1-score (W-). In each population group, the W- values in bold represent the best performances achieved for each feature set considered, while in red is highlighted the best accuracy at all.
| | PPG Features | GSR Features | PPG and GSR Features |
---|
| Classifier | | MC | AL | | | MC | AL | | | MC | AL | |
Acc. | | | W- | Acc. | | | W- | Acc. | | | W- |
Young Adults | SVM Linear | 79% | 0.77 | 0.80 | 79% | 93% | 0.93 | 0.94 | 93% | 91% | 0.91 | 0.91 | 91% |
SVM Cubic | 72% | 0.72 | 0.72 | 72% | 91% | 0.91 | 0.92 | 91% | 80% | 0.80 | 0.79 | 80% |
SVM Gauss | 76% | 0.75 | 0.76 | 76% | 91% | 0.90 | 0.91 | 91% | 83% | 0.84 | 0.83 | 83% |
Cart | 64% | 0.63 | 0.64 | 64% | 93% | 0.93 | 0.93 | 93% | 91% | 0.91 | 0.91 | 91% |
Random | 50% | 0.50 | 0.50 | 50% | 50% | 0.50 | 0.50 | 50% | 50% | 0.50 | 0.50 | 50% |
Older Adults | SVM Linear | 80% | 0.80 | 0.81 | 80% | 92% | 0.92 | 0.92 | 92% | 90% | 0.90 | 0.90 | 90% |
SVM Cubic | 72% | 0.71 | 0.73 | 72% | 90% | 0.90 | 0.91 | 90% | 89% | 0.89 | 0.89 | 89% |
SVM Gauss | 78% | 0.78 | 0.78 | 78% | 92% | 0.92 | 0.92 | 92% | 89% | 0.89 | 0.89 | 89% |
Cart | 75% | 0.75 | 0.74 | 75% | 88% | 0.88 | 0.89 | 88% | 90% | 0.90 | 0.90 | 90% |
Random | 50% | 0.50 | 0.50 | 50% | 50% | 0.50 | 0.50 | 50% | 50% | 0.50 | 0.50 | 50% |
Table 5.
Percentage of instances for each cognitive load task classified as high arousal and low arousal using pre-trained SVM-Linear classifiers. Each classifier has been trained considering the instances collected during the MC task as high arousal and the instances collected from AL as low arousal. The classifiers are then applied to classify the following: baseline task B, reading task R and comprehension task C. In the first three rows, the results generated using the signals collected from young adults (Yng) are reported, while the last three rows are related to signals acquired from the older adults participants (Old). In each population group and cognitive load task, the values in bold represent the highest percentage of instances between high and low arousal for each feature set considered.
Table 5.
Percentage of instances for each cognitive load task classified as high arousal and low arousal using pre-trained SVM-Linear classifiers. Each classifier has been trained considering the instances collected during the MC task as high arousal and the instances collected from AL as low arousal. The classifiers are then applied to classify the following: baseline task B, reading task R and comprehension task C. In the first three rows, the results generated using the signals collected from young adults (Yng) are reported, while the last three rows are related to signals acquired from the older adults participants (Old). In each population group and cognitive load task, the values in bold represent the highest percentage of instances between high and low arousal for each feature set considered.
| | PPG | GSR | PPG + GSR |
---|
| | % InstancesHigh Arousal | % InstancesLow Arousal | % InstancesHigh Arousal | % InstancesLow Arousal | % InstancesHigh Arousal | % InstancesLow Arousal |
Yng | B | 17% | 83% | 17% | 83% | 17% | 83% |
R | 34% | 66% | 28% | 72% | 22% | 78% |
C | 69% | 31% | 84% | 16% | 75% | 25% |
Old | B | 7% | 93% | 10% | 90% | 8% | 93% |
R | 50% | 50% | 35% | 65% | 40% | 60% |
C | 53% | 47% | 83% | 17% | 73% | 27% |
Table 6.
The frequency strides estimated for both subject groups are reported and compared with the metronome frequencies (F1, F2 and F3).
Table 6.
The frequency strides estimated for both subject groups are reported and compared with the metronome frequencies (F1, F2 and F3).
| Young Adults | Older Adults |
---|
Metronome | EMG | EMG |
F1 = 0.58 | 0.59 | 0.66 |
F2 = 0.70 | 0.72 | 0.76 |
F3 = 0.83 | 0.85 | 0.85 |
Table 7.
Kruskal Wallis p-values: comparison between tasks in young adults. The values highlighted in red refer to p-value lower than the significance level chosen: . The analyzed tasks are: B = baseline task acquired in walking session, F1 = metronome forced speed task (70 bpm), F2 = metronome forced speed task (85 bpm), F3 = metronome forced speed task (100 bpm), WO + Obs = single signal for the whole task of collision avoidance (free walk and obstacle crossing).
Table 7.
Kruskal Wallis p-values: comparison between tasks in young adults. The values highlighted in red refer to p-value lower than the significance level chosen: . The analyzed tasks are: B = baseline task acquired in walking session, F1 = metronome forced speed task (70 bpm), F2 = metronome forced speed task (85 bpm), F3 = metronome forced speed task (100 bpm), WO + Obs = single signal for the whole task of collision avoidance (free walk and obstacle crossing).
First Task | Second Task | Maximum | Minimum | Mean | Variance | Peak Rate | IBI | RMSSD |
---|
B | F1 | <0.001 | <0.001 | 0.29 | <0.001 | <0.001 | <0.001 | <0.001 |
B | F2 | <0.001 | <0.001 | 0.21 | <0.001 | 0.24 | 0.45 | <0.001 |
B | F3 | <0.001 | <0.001 | 0.03 | <0.001 | <0.001 | <0.001 | <0.001 |
B | WO + Obs | <0.001 | <0.001 | 0.01 | <0.001 | <0.001 | <0.001 | <0.001 |
F1 | F2 | 0.58 | 0.66 | 0.80 | 0.64 | <0.001 | <0.001 | 0.14 |
F1 | F3 | 0.77 | 0.64 | 0.28 | 0.14 | <0.001 | <0.001 | 0.70 |
F1 | WO + Obs | <0.001 | <0.001 | 0.18 | <0.001 | <0.001 | <0.001 | <0.001 |
F2 | F3 | 0.51 | 0.91 | 0.42 | 0.37 | <0.001 | <0.001 | 0.08 |
F2 | WO + Obs | <0.001 | <0.001 | 0.20 | 0.01 | <0.001 | <0.001 | 0.08 |
F3 | WO + Obs | <0.001 | <0.001 | 0.71 | 0.17 | 0.10 | 0.24 | <0.001 |
Table 8.
Kruskal Wallis p-values: comparison between tasks in older adults. The values highlighted in red refer to p-values lower than the significance level . The analyzed tasks are: B = baseline collected during walking session F1 = metronome forced speed task (70 bpm), F2 = metronome forced speed task (85 bpm), F3 = metronome forced speed task (100 bpm), FW = pure free walk task, WO = free walk in the collision avoidance task, Obs = obstacle crossing.
Table 8.
Kruskal Wallis p-values: comparison between tasks in older adults. The values highlighted in red refer to p-values lower than the significance level . The analyzed tasks are: B = baseline collected during walking session F1 = metronome forced speed task (70 bpm), F2 = metronome forced speed task (85 bpm), F3 = metronome forced speed task (100 bpm), FW = pure free walk task, WO = free walk in the collision avoidance task, Obs = obstacle crossing.
First Task | Second Task | Maximum | Minimum | Mean | Variance | Peak Rate | IBI | RMSSD |
---|
B | F1 | <0.001 | <0.001 | <0.001 | <0.001 | 0.18 | 0.15 | <0.001 |
B | F2 | <0.001 | <0.001 | <0.001 | <0.001 | <0.001 | <0.001 | <0.001 |
B | F3 | <0.001 | <0.001 | <0.001 | <0.001 | <0.001 | <0.001 | <0.001 |
B | FW | <0.001 | <0.001 | 0.20 | <0.001 | <0.001 | <0.001 | <0.001 |
B | Obs | 0.19 | <0.001 | 0.01 | <0.001 | <0.001 | <0.001 | <0.001 |
B | WO | <0.001 | <0.001 | <0.001 | <0.001 | <0.001 | <0.001 | <0.001 |
F1 | F2 | 0.78 | 0.89 | 0.98 | 0.66 | 0.03 | 0.03 | 0.56 |
F1 | F3 | 0.90 | 0.60 | 0.73 | 0.49 | <0.001 | <0.001 | 0.30 |
F1 | FW | 0.28 | <0.001 | 0.01 | 0.78 | <0.001 | <0.001 | 0.13 |
F1 | Obs | 0.07 | 0.49 | 0.88 | 0.10 | <0.001 | <0.001 | 0.13 |
F1 | WO | 0.38 | 0.74 | 0.01 | 0.08 | <0.001 | <0.001 | 0.31 |
F2 | F3 | 0.90 | 0.64 | 0.82 | 0.73 | 0.16 | 0.18 | 0.58 |
F2 | FW | 0.38 | 0.01 | 0.01 | 0.82 | 0.38 | 0.27 | 0.30 |
F2 | Obs | 0.04 | 0.32 | 0.99 | 0.19 | <0.001 | 0.02 | 0.41 |
F2 | WO | 0.30 | 0.85 | 0.02 | 0.15 | <0.001 | 0.02 | 0.77 |
F3 | FW | 0.25 | 0.02 | <0.001 | 0.60 | 0.67 | 0.84 | 0.73 |
F3 | Obs | 0.07 | 0.19 | 0.92 | 0.35 | <0.001 | 0.23 | 0.91 |
F3 | WO | 0.36 | 0.82 | 0.04 | 0.33 | 0.03 | 0.36 | 0.66 |
FW | Obs | 0.01 | <0.001 | 0.05 | 0.14 | <0.001 | 0.14 | 0.77 |
FW | WO | 0.06 | <0.001 | <0.001 | 0.09 | 0.01 | 0.20 | 0.35 |
Obs | WO | 0.18 | 0.18 | 0.05 | 0.94 | 0.01 | 0.72 | 0.51 |
Table 9.
Performance of the binary classifiers in discriminating walking tasks having similar walking pace considering the signals collected from older adults. Three tasks are compared: pure free walk FW, free walk before and after the collision avoidance zone WO and obstacle crossing Obs. The analysis are performed varying the feature set used (PPG, GSR or PPG and GSR) and adopting a LOSO validation strategy. Three performance metrics are reported: accuracy (Acc.), F1-score () and weighted F1-score (W-). The W- values in bold represent the best performances achieved for each feature set considered, while in red is highlighted the best accuracy at all.
Table 9.
Performance of the binary classifiers in discriminating walking tasks having similar walking pace considering the signals collected from older adults. Three tasks are compared: pure free walk FW, free walk before and after the collision avoidance zone WO and obstacle crossing Obs. The analysis are performed varying the feature set used (PPG, GSR or PPG and GSR) and adopting a LOSO validation strategy. Three performance metrics are reported: accuracy (Acc.), F1-score () and weighted F1-score (W-). The W- values in bold represent the best performances achieved for each feature set considered, while in red is highlighted the best accuracy at all.
| | WO vs. FW | FW vs. Obs | WO vs. Obs |
---|
| Classifier | | WO | FW | | | Obs | FW | | | Obs | WO | |
Acc. | | | W- | Acc. | | | W- | Acc. | | | W- |
PPG | SVM Linear | 68% | 0.73 | 0.61 | 68% | 83% | 0.80 | 0.85 | 83% | 66% | 0.45 | 0.75 | 63% |
SVM Cubic | 70% | 0.75 | 0.62 | 70% | 70% | 0.67 | 0.72 | 70% | 55% | 0.37 | 0.65 | 54% |
SVM Gauss | 67% | 0.74 | 0.54 | 66% | 76% | 0.71 | 0.80 | 76% | 64% | 0.32 | 0.75 | 58% |
Cart | 64% | 0.71 | 0.52 | 63% | 71% | 0.69 | 0.72 | 71% | 55% | 0.40 | 0.64 | 54% |
Random | 50% | 0.54 | 0.45 | 50% | 50% | 0.51 | 0.49 | 50% | 50% | 0.55 | 0.44 | 51% |
GSR | SVM Linear | 64% | 0.70 | 0.53 | 63% | 77% | 0.74 | 0.79 | 77% | 65% | 0.30 | 0.76 | 58% |
SVM Cubic | 58% | 0.62 | 0.54 | 59% | 80% | 0.79 | 0.82 | 80% | 64% | 0.50 | 0.72 | 63% |
SVM Gauss | 62% | 0.69 | 0.51 | 62% | 81% | 0.79 | 0.82 | 81% | 66% | 0.44 | 0.76 | 63% |
Cart | 63% | 0.69 | 0.56 | 63% | 74% | 0.73 | 0.74 | 74% | 58% | 0.47 | 0.65 | 58% |
Random | 50% | 0.54 | 0.45 | 50% | 50% | 0.51 | 0.49 | 50% | 50% | 0.55 | 0.44 | 51% |
PPG and GSR | SVM Linear | 71% | 0.75 | 0.66 | 71% | 81% | 0.78 | 0.83 | 80% | 67% | 0.46 | 0.77 | 65% |
SVM Cubic | 66% | 0.71 | 0.59 | 66% | 80% | 0.78 | 0.82 | 80% | 64% | 0.51 | 0.71 | 63% |
SVM Gauss | 73% | 0.77 | 0.65 | 72% | 81% | 0.80 | 0.83 | 81% | 68% | 0.52 | 0.76 | 66% |
Cart | 64% | 0.71 | 0.53 | 63% | 69% | 0.67 | 0.71 | 69% | 68% | 0.56 | 0.74 | 67% |
Random | 50% | 0.54 | 0.45 | 50% | 50% | 0.51 | 0.49 | 50% | 50% | 0.55 | 0.44 | 51% |
Table 10.
Performance of the binary classifiers in discriminating tasks having similar walking pace considering the signals collected from young adults. Two tasks have been compared: the metronome forced speed task of 100 bpm F3 and the collision avoidance task WO+Obs, including, in a single instance, both the free walking before and after the collision avoidance zone and the obstacle crossing. The analysis are performed varying the feature set used (PPG, GSR or PPG and GSR) and adopting a LOSO validation strategy. Three performance metrics are reported: accuracy (Acc.), F1-score () and weighted F1-score (W-). The W- values in bold represent the best performances achieved for each feature set considered, while in red is highlighted the best accuracy at all.
Table 10.
Performance of the binary classifiers in discriminating tasks having similar walking pace considering the signals collected from young adults. Two tasks have been compared: the metronome forced speed task of 100 bpm F3 and the collision avoidance task WO+Obs, including, in a single instance, both the free walking before and after the collision avoidance zone and the obstacle crossing. The analysis are performed varying the feature set used (PPG, GSR or PPG and GSR) and adopting a LOSO validation strategy. Three performance metrics are reported: accuracy (Acc.), F1-score () and weighted F1-score (W-). The W- values in bold represent the best performances achieved for each feature set considered, while in red is highlighted the best accuracy at all.
| PPG Features | GSR Features | PPG and GSR Features |
---|
Classifier | | F3 | WO + Obs | | | F3 | WO + Obs | | | F3 | WO + Obs | |
Acc. | | | W- | Acc. | | | W- | Acc. | | | W- |
SVM Linear | 74% | 0.76 | 0.71 | 74% | 65% | 0.61 | 0.69 | 65% | 72% | 0.72 | 0.71 | 72% |
SVM Cubic | 60% | 0.62 | 0.57 | 60% | 65% | 0.67 | 0.64 | 65% | 62% | 0.65 | 0.59 | 62% |
SVM Gauss | 72% | 0.75 | 0.68 | 71% | 66% | 0.61 | 0.70 | 66% | 68% | 0.69 | 0.68 | 68% |
Cart | 63% | 0.67 | 0.58 | 62% | 63% | 0.67 | 0.59 | 63% | 62% | 0.65 | 0.58 | 62% |
Random | 50% | 0.50 | 0.50 | 50% | 50% | 0.50 | 0.50 | 50% | 50% | 0.50 | 0.50 | 50% |