Table 4. Confusion Matrix of the Time Spent Performing Activities with Triaxial Accelerometers when using six and eight Features.
| CE | UE f | ||||
|---|---|---|---|---|---|
| 6 features | 8 features | 6 features | 8 features | ||
| Sensitivity (%) a | Walking | 91±7 | 91±7 | 83±6 | 83±6 |
| Sitting | 92±8 | 94±7 | 93±3 | 94±5 | |
| Running | 64±25 | 64±25 | - | - | |
| Lying Down | 92±4 | 90±11 | 88±11 | 87±10A | |
| Dynamic Standing | 85±11 | 87±9 | 79±8 | 79±8 | |
| F-Score (%) b | Walking | 85±6 | 85±6 | 77±4 | 77±4 |
| Sitting | 94±5 | 95±3 | 90±5 | 90±5 | |
| Running | 69±21 | 69±21 | - | - | |
| Lying Down | 86±13 | 88±9 | 91±6 | 91±5A | |
| Dynamic Standing | 88±6 | 89±5 | 83±5 | 84±5 | |
| Specificity (%) c | Walking | 94±3 | 94±3 | 98±1 | 98±1 |
| Sitting | 98±1 | 97±2 | 90±5 | 89±7 | |
| Running | 99±2 | 99±2 | 100±0 | 100±0 | |
| Lying Down | 96±5 | 97±3 | 97±2 | 97±4 | |
| Dynamic Standing | 99±1 | 99±1 | 99±1 | 99±1 | |
| Overall Accuracy (%) d | 88±6 | 89±4 | 89±4 | 89±4 | |
| Kappa e | 0.84±0.01 | 0.85±0.00 | 0.82±0.00 | 0.83±0.00 | |
| Kappa linear weight | 0.81±0.01 | 0.81±0.00 | 0.79±0.00 | 0.80±0.00 | |
| Kappa quadratic weight | 0.79±0.02 | 0.79±0.02 | 0.76±0.01 | 0.76±0.01 | |
a Sensitivity corresponds to the chances of classifying an activity as positive when it is indeed positive
b F-Score is defined as the "harmonic mean between sensitivity and positive predictive values" [18]
c Specificity is a measure of chances of classifying an activity as negative when they are truly negative
d Overall accuracy is the mean proportion of all activities that are correctly classified per person
e Cohen’s Kappa is the measure of the agreement between the real activity and the classifications
f In UE, the results for running are not included since no participants had practiced this activity and the results for lying down are based on 5 participants since one participant had not practiced this activity.