Table 4 Performance metrics of the best performing classifier (‘fair’ case) when the weight is used as a criterion to select the subjects of the training subset.
Dataset | Features and algorithm | Subjects included in the training subset | Se (%) | Sp (%) | \(\sqrt{Se\cdot Sp}\) (%) |
|---|---|---|---|---|---|
DOFDA | HCTSA features Naive Bayes (Gaussian) | Random selection of users | 97.38 | 100.00 | 98.67 |
All (fair distribution) | 97.37 | 100.00 | 98.67 | ||
Subjects (80%) with highest weight | 94.81 | 100.00 | 97.37 | ||
Subjects (80%) with lowest weight | 93.51 | 100.00 | 96.70 | ||
Erciyes | Own selection of features SVM (quadratic kernel) | Subjects (80%) with highest weight | 100.00 | 100.00 | 100.00 |
All (fair distribution) | 99.62 | 99.18 | 99.40 | ||
Subjects (80%) with lowest weight | 99.41 | 97.06 | 98.23 | ||
Random selection of users | 97.83 | 98.43 | 98.12 | ||
SisFall | HCTSA features SVM (cubic kernel) | Subjects (80%) with highest weight | 100.00 | 100.00 | 100.00 |
All (fair distribution) | 99.78 | 99.96 | 99.87 | ||
Random selection of users | 99.74 | 99.96 | 99.85 | ||
Subjects (80%) with lowest weight | 85.33 | 100.00 | 92.38 | ||
UMAFall | Own selection of features KNN (Euclidean. 10 neighbors) | All (fair distribution) | 98.93 | 98.73 | 98.83 |
Subjects (80%) with lowest weight | 100.00 | 95.38 | 97.67 | ||
Random selection of users | 98.28 | 97.05 | 97.66 | ||
Subjects (80%) with highest weight | 91.55 | 98.68 | 95.05 | ||
UP-Fall | Own selection of features SVM (linear kernel) | Subjects (80%) with highest weight | 100.00 | 100.00 | 100.00 |
All (fair distribution) | 99.59 | 98.02 | 98.80 | ||
Subjects (80%) with lowest weight | 100.00 | 97.56 | 98.77 | ||
Random selection of users | 99.65 | 97.56 | 98.60 |
