Table 3 Performance metrics of the four best performing classifiers using fivefold cross validation and a ‘fair’ distribution of the samples between the training and testing subsets.
Dataset | Features | Algorithm and hyperparameters | Se (%) | Sp (%) | \(\sqrt{Se\cdot Sp}\)a |
|---|---|---|---|---|---|
DOFDA | HCTSA | Naive Bayes (Gaussian) | 97.37 | 100.00 | 98.67 ± 1.27% |
HCTSA | SVM (linear kernel) | 99.01 | 98.33 | 98.65 ± 1.77% | |
Own selection | KNN (Euclidean, 10 neighbors) | 98.02 | 98.18 | 98.08 ± 2.08% | |
HCTSA | SVM (quadratic kernel) | 99.34 | 96.67 | 97.97 ± 2.44% | |
Erciyes | Own selection | SVM (quadratic kernel) | 99.62 | 99.18 | 99.40 ± 0.17% |
Own selection | SVM (medium gaussian kernel) | 99.34 | 98.98 | 99.16 ± 0.17% | |
Own selection | KNN (cosine, 5 neighbors) | 99.07 | 99.05 | 99.06 ± 0.12% | |
Own selection | KNN (Minkowski, 5 neighbors) | 99.45 | 98.64 | 99.04 ± 0.12% | |
SisFall | HCTSA | SVM (cubic kernel) | 99.78 | 99.96 | 99.87 ± 0.13% |
HCTSA | SVM (quadratic kernel) | 99.78 | 99.96 | 99.87 ± 0.19% | |
HCTSA | SVM (medium gaussian kernel) | 99.11 | 99.96 | 99.54 ± 0.12% | |
HCTSA | DT (Fine) | 98.89 | 99.96 | 99.42 ± 0.23% | |
UMAFall | Own selection | KNN (Euclidean, 10 neighbors) | 98.93 | 98.73 | 98.83 ± 0.86% |
Own selection | DT (Coarse model) | 98.38 | 98.99 | 98.67 ± 1.93% | |
Own selection | SVM (medium gaussian kernel) | 97.87 | 99.24 | 98.55 ± 0.55% | |
Own selection | KNN (Euclidean, 5 neighbors) | 98.93 | 97.97 | 98.45 ± 1.10% | |
UP-Fall | Own selection | SVM (linear kernel) | 99.59 | 98.02 | 98.80 ± 1.31% |
Own selection | SVM (medium gaussian kernel) | 98.78 | 98.82 | 98.79 ± 1.32% | |
Own selection | KNN (Euclidean, 10 neighbors) | 99.18 | 97.23 | 98.20 ± 1.31% | |
Own selection | KNN (Euclidean, 5 neighbors) | 98.78 | 97.62 | 98.19 ± 1.65% |
