Table 5.
Summary of characteristic of studies that investigated drowsiness quantification using wearable devices.
| No. | Type of study | # of participants | Fatiguing task | Task duration | Input data | Reference measure | Modeling approach | Output | Model performance |
|---|---|---|---|---|---|---|---|---|---|
| 7 | Lab | 28 | Simulated monotonous driving | 120 min | GSR, PPG, TSk, motion | Video-based reference | SVM | 4 states (normal, stressed, fatigued and/or drowsy) | ACC = 68.31% (4 states)/ ACC = 84.46% (3 states) |
| 12 | Lab | 15 | Simulated night-time highway driving (lane-departure paradigm) | 60 min | EEG (brain activity and eye blinking) | Reaction time | Multiple Linear regression | Binary | Se = 58% Sp = 73% ACC = 68% |
| 13 | Lab | 10 | Simulated highway driving | Not stated | Motion | Video-based reference, KSS | SVM | Binary | ACC = 83.3% |
| 14 | Lab | 29 | Target hitting game (alertness activity) and simulated driving | 7 min + 60 min | EEG | KSS | Iterative negative-unlabelled learning algorithm | Subject's most fatigued block | ACC = 93.8% |
| 15 | Lab | 10 | Simulated driving | 90 min | Motion | Video-based reference | SVM | Binary | ACC = 82.6% |
| 16 | Lab | 20 | Simulated driving | 120 min | Eye movement | Real fatigue probability calculated based on heart rate test and subjective evaluation | HMM | Binary | ACC = 80% |
| 17 | Lab | 10 | Simulated train driving while sleep deprived | Not stated | EEG | Investigator's observation | SVM | Binary | ACC = 90.7% Se = 86.8% FP = 5.4% |
| 18 | Field | 12 | Real highway driving | 210 min | EEG, EMG, RES, context | Not stated | First order HMM | Probabilities of fatigue | AUC = 0.84* |
| 19 | Lab | 6 | Not stated | Not stated | EMG, GSR | KSS, physician observation | SVM | Binary | Precision rate = 92% |
| 20 | Lab | 15 | Simulated driving | 60 min | EEG | Reaction time | Linear regression model | Reaction time | ACC = 93.9% RMSE = 219.98 ms |
| 21 | Lab | 28 | Level-2 automated ride | 45 min | PPG | KSS | KNN | Binary/3 levels | ACC = 99.4% F-score = 0.99/ ACC = 98.5% F-score = 0.99 |
| 22 | Lab | 27 | Simulated automated ride | 45 min | PPG | Weinbeer scale, micro-sleep events (based on eye closure duration) | Decision Stump | Binary | ACC = 73.4% F-score = 0.74** |
| 23 | Lab | 10 (study A)/30 (study B) | Simulated monotonous driving under sleep deprivation/simulated monotonous driving | 60 min/45 min | PPG | KSS and video-based reference/KSS | Subspace KNN | Binary | ACC = 99.9% Se = 100% Sp = 100% precision = 100% NPV = 100% |
| 24 | Lab | 19 | Simulated monotonous driving under different lightning conditions and levels of communication between subjects and researcher | 90–150 min | TSk | SSS | Decision Trees | Binary | Se = 77.8% Sp = 100% ACC = 88.9% |
| 25 | Lab | 50 | Watching a 3D rotating screen saver while sitting on a comfortable seat and sleep deprived | 20 min | EEG (brain activity and eye blinking), motion | KSS | SVM with linear kernel | Binary | ACC = 86.5% (LOOCV)/ ACC = 92% Se = 88% Sp = 96% precision = 95.6% (Hold-out validation) |
| 26 | Lab | 6 | Simulated driving | 60 min | ECG | Video-based reference | SVM regression | Binary | AUC = 0.95 |
| 27 | Lab | 3 | Simulating drowsy state in the late night | 5 times per state (approx. 4 min in total) | EEG (brain activity and eye blinking), motion | Parameters threshold determined by authors | Nearest Centroid Classifier based on K-means clustering | 5 levels | ACC = 83% |
| 28 | Lab | 6 | Simulated driving | 60–120 min | ECG/PPG | Video-based reference | Convolutional neural network | Binary | ACC = 70%/64% precision = 71%/71% recall = 85%/78% F-score = 77%/71% |
| 29 | Lab | 4 | Hand-eye-coordination game in a sleep deprived condition | approx. 8 min (500 s) | EEG | Not stated | Random forest | 3 states (normal, sleepy, fallen) | ACC = 98% |
| 30 | Field | 29 | Not stated | Not stated | EEG | KSS | SVM with radial basis function kernel | Binary | precision = 73.5% Se = 88.7% Sp = 45.2% ACC = 72.7% F-score = 80.4% |
| 31 | Lab | 23 | 4 naps while sleep deprived | 20 min each nap | EEG | Clinician scoring (based on EEG data) | SVM with radial basis function kernel | Binary | ACC = 80% Kappa coefficient = 0.53 |
| 32 | Lab | 10 | Simulated high-speed train driving while sleep deprived | Not stated | EEG | Not stated | Robust principal component analysis algorithm | Binary | ACC = 99.4% |
| 33 | Lab | 15 | Simulated driving | 60 min | PPG | KSS, reaction time, total overrun area | KNN | Binary/3 levels | ACC = 93%/75% |
| 34 | Lab | 17 | Simulated monotonous driving | 60 min | EEG, motion | Wierwille scale | linear SVM | Binary/5 levels | ACC = 96.2%/93.7% |
| 35 | Lab | 20 | Simulated monotonous driving | 60 min | EEG | PERCLOS, number of adjustments on steering wheel | SVM-based posterior probabilistic model | 3 levels (alert, early warning, full warning) | ACC = 89%*** |
| 36 | Lab | 6 | Simulated monotonous driving | 60 min | EEG, motion | Wierwille scale | SVM with linear kernel | Binary | ACC = 96.2% Se = 96.5% Sp = 95.6% |
| 37 | Lab | 20 | Simulated driving | 60 min | Motion | KSS (rated by observer and confirmed by participant) | SVM | 5 levels | ACC = 98.2% |
| 38 | Lab | 20 | Simulated driving | 60 min | PPG, GSR, motion | KSS (rated by observer and confirmed by participant) | SVM | 5 levels | ACC = 98.3% |
| 39 | Lab | 4 | Simulated driving | Not stated | PPG, motion | Video-based reference, driver's physical state | SVM with radial basis function kernel | Binary | ACC = 94.4% precision-recall score = 96.4% |
| 40 | Lab | 10 | Watching the photographed actual road video before and after doing a PVT | 50 min | PPG | Not stated | SVM | Binary | ACC = 96.3% recall = 94.7% precision = 97.8% |
| 41 | Lab | 23 | Simulated driving while sleep deprived | 60 min | EEG | Not stated | SVM with temporal aggregation | Binary | ACC = 87% |
| 42 | Lab | 45 | Sit on a chair and watch a movie of night driving while holding a steering wheel | 80 min | EOG | Video-based reference | Random forest | Binary | ACC = 80% |
| 43 | Lab | Unclear | Sleep deprivation (4 h of sleep) | 20 min | EEG, NIRS | Oxford Sleep Resistance Test | 3rd order polynomial SVM | 3 levels (1, 2, or 3 consecutive missed stimulus) | ACC = 77.3%**** |
| 44 | Lab | 12 | Simulated driving | 480 min | PPG, motion | Observation, KSS (rated by observer) | Mobile-based SVM | Binary | ACC = 95.8% |
ECG,electroencephalogram; EEG,electroencephalogram; EMG,electromyogram; EOG,electrooculogram; GSR,galvanic skin response; NIRS,near-infra-red spectroscopy; PPG,photoplethysmogram; RES,respiration; TSk,skin temperature; PVT,psychomotor vigilance test; KSS,Karolinska Sleepiness Scale; PERCLOS,percentage of eye closure; SSS,Stanford Sleepiness Scale; KNN,k-nearest neighbors; HMM,hidden Markov model; SVM,Support Vector Machine; ACC,accuracy; AUC,area under the receiver operating characteristic curve; FP,false positive; LOOCV,leave-one-out cross-validation; NPV,negative predictive value; RMSE,root mean square error; Se,sensitivity; Sp,specificity.
*
Calculated average among all subjects (AUC between 0.734 and 0.960).
**
Calculated average of classes (0.82 for non-drowsy and 0.65 drowsy class).
***
Average of the 3 states (91.25% for alert, 83.8% for early-warning group, and 91.9% for full warning group).
****
Calculated average of classes (88.1 for 1, 77.9 for 2, and 65.9 for 3 missed stimulus).