Table 9 Performance metrics of the best performing classifier (‘fair’ case) when different categories of ADL are used in the training and the testing subsets.
Dataset | Features and algorithm | ADL categories used for | Se (%) | Sp (%) | \(\sqrt{Se\cdot Sp}\) (%) | |
|---|---|---|---|---|---|---|
Training | Test | |||||
DOFDA | HCTSA features Naïve Bayes (Gaussian) | All (fair distribution) | All (fair distribution) | 97.37 | 100.00 | 98.67 |
Basic ADLs | Standard ADLs | 100.00 | 68.00 | 82.46 | ||
Standard ADLs | Basic ADLs | 100.00 | 50.00 | 70.71 | ||
Erciyes | Own selection of features SVM (quadratic kernel) | All (fair distribution) | All (fair distribution) | 99.62 | 99.18 | 99.40 |
All but standard ADLs | Standard ADLs | 99.34 | 98.90 | 99.12 | ||
All but basic ADLs | Basic ADLs | 99.34 | 97.22 | 98.28 | ||
All but sporting ADLs | Sporting ADLs | 98.68 | 95.65 | 97.15 | ||
All but ‘Near Falls’ | Near Falls | 100.00 | 92.39 | 96.12 | ||
SisFall | HCTSA SVM (cubic kernel) | All (fair distribution) | All (fair distribution) | 99.78 | 99.96 | 99.87 |
All but basic ADLs | Basic ADLs | 99.83 | 92.96 | 96.34 | ||
All but sporting ADLs | Sporting ADLs | 99.67 | 84.46 | 91.75 | ||
All but standard ADLs | Standard ADLs | 99.67 | 72.34 | 84.91 | ||
UMAFall | Own selection of features KNN (Euclidean. 10 neighbors) | All (fair distribution) | All (fair distribution) | 98.93 | 98.73 | 98.83 |
All but basic ADLs | Basic ADLs | 100.00 | 93.84 | 96.87 | ||
Standard ADLs | Standard ADLs | 95.16 | 97.87 | 96.51 | ||
All but sporting ADLs | Sporting ADLs | 100.00 | 1.82 | 13.48 | ||
UP-Fall | Own selection of features SVM (linear kernel) | All (fair distribution) | All (fair distribution) | 99.59 | 98.02 | 98.80 |
All but basic ADLs | Basic ADLs | 100.00 | 100.00 | 100.00 | ||
Standard ADLs | Standard ADLs | 98.77 | 95.93 | 97.34 | ||
All but sporting ADLs | Sporting ADLs | 100.00 | 2.17 | 14.74 | ||
