Table 16.
Comparison with other accelerometer-based phone placement solutions. LOSO, leave-one-subject-out.
| Park et al. [19] | Wiese et al. [8] | Fujinami et al. [20] | Alanezi et al. [22] | Wahl et al. [25] | Kunze et al. [9] | Mannini et al. [18] | This Work | |
|---|---|---|---|---|---|---|---|---|
| Positions | 4: Bag, ear, hand, pocket | 4: Pocket, bag, hand, out | 9: Neck, chest, jacket pocket, trousers pockets (front, back), backpack, handbag, messenger bag, shoulder bag | 6: hand-holding, talking on phone, watching a video, pockets (pants, hip, jacket) | 5: Pants, table, jacket, bag, no label | 5: Head, wrist, torso pockets (front, back); AND 5: hand, wrist upper arm, knee and back | 5: Ankle, thigh, hip, upper arm, wrist | 8: Backpack, hand, pocket, messenger-bag, jacket-pocket, arm, belt, wrist |
| Activities | Walking | sitting (on a couch, on a desk chair), standing, walking | Walking | Idle, walking, running | Working, eating, walking/cycling, Vehicle | activities of daily living, household, workshop and office activities | Lying (on back, on left side, on right side), sitting (Internet searching, typing, writing, reading), standing still, sorting files on paperwork, exercise bike, cycling (outdoor level, outdoor uphill, outdoor downhill), elevator (up, down), jumping jacks, sweeping with broom, painting with roller, painting with brush, walking, stairs | Sit, stand, bike, walk, run, jog, stairs (up/down), transport (bus), secondary activities: sending an SMS, making a call, interaction with an app |
| No. of Subjects | 14 | 15–32 | 20 | 10 | 6 | 17 | 33 | 35 |
| Features | Frequency domain | Time and frequency domain | Time and frequency domain | Time domain | Time domain | Time domain | Time and frequency domain | Time and frequency domain |
| Classifier | C4.5, SVM | SVM, random forest | J48, SVM, naive Bayes, MLP | J48, naive Bayes, logistic regression, MLP | Nearest centroid classifier | HMM + particle filter smoothing | SVM | J48, KNN, random forest, MLP |
| Validation method | 10-fold cross | LOSO | LOSO | 10-fold cross | 10-fold cross | Train/test over randomly-picked subset | LOSO | LOSO |
| Recognition accuracy | 99.6% | 79% (random forest) | 74.6% | 88.5% (only accelerometer, first activity detected, J48) | 82% (only accelerometer) | 82.0% | 92.7% (first walking activity detected) | 85% (datasets combined) |
| Recognition accuracy, walking | 99.6% | Not reported | 74.6% | Not reported | Not reported | 94.9% | 81% | 88% (datasets combined) |