Performance of Random Forest algorithms for four difference cases in three phone locations; right hand, right pocketand backpack
| Accuracy | ROC | PR-AUC | Sensitivity | Specificity | |
| Right hand | |||||
| Case 1 | 66.8 | 0.91 | 0.71 | 0.65 | 0.93 |
| Case 2 | 40.3 | 0.72 | 0.34 | 0.38 | 0.88 |
| Case 3 | 89 | 0.99 | 0.95 | 0.89 | 0.98 |
| Case 4 | 48.5 | 0.83 | 0.52 | 0.46 | 0.90 |
| Right pocket | |||||
| Case 1 | 67.2 | 0.91 | 0.69 | 0.65 | 0.94 |
| Case 2 | 39.7 | 0.74 | 0.34 | 0.38 | 0.88 |
| Case 3 | 92.9 | 0.99 | 0.97 | 0.93 | 0.99 |
| Case 4 | 51.4 | 0.986 | 0.857 | 0.49 | 0.90 |
| Backpack | |||||
| Case 1 | 69.3 | 0.92 | 0.72 | 0.68 | 0.94 |
| Case 2 | 40.4 | 0.74 | 0.37 | 0.39 | 0.88 |
| Case 3 | 90.8 | 0.99 | 0.96 | 0.90 | 0.98 |
| Case 4 | 52.1 | 0.86 | 0.58 | 0.50 | 0.90 |
Case 1 features include only raw X, Y, Z acceleration. Case 2 feature includes only the vector magnitude of activity counts. Case 3 includes raw accelerometer data and 58 features. Case 4 includes the vector magnitude of activity counts and 58 features.
PR-AUC, area under the Precision-Recall curve; ROC, receiver operating characteristic.