Table 7.

Vehicle-based drowsiness detection systems.

Ref.	Vehicle Parameters	Extracted Features	Classification Method	Description	Quality Metric	Dataset
[113]	Steering wheel	SWA	RF	Used SWA as input data and compared it with PERCLOS. The RF algorithm was trained by a series of decision trees, with a randomly selected feature.	Accuracy: RF- steering model: 79% PERCLOS: 55%	Prepared their own dataset
[114]	Lateral distance	Statistical features, derived from the time and wavelet domains, relevant to the lateral distance and lane trajectory	SVM and neural network	Detection was based on lateral distance. Additionally, it collects data of the driver’s facial and head movements to be used as ground truth for the vehicle data.	Accuracy: Over 90%	Prepared their own dataset
[117]	Steering wheel	SWA	Specially designed binary decision classifier	Used SWA data to apply online fatigue detection. The alertness state is determined using a specially designed classifier.	Accuracy: Drowsy: 84.85% Awake: 78.01%	Prepared their own dataset
[118]	Steering wheel	SWA, steering wheel velocity	ANFIS for feature selection, PSO for optimizing the ANFIS parameters, and SVM for classification	Detection was based on steering wheel data. The system used a selection method that utilized ANFIS.	Accuracy: 98.12%	Prepared their own dataset
[119]	Steering wheel	SW_Range_2, Amp_D2_Theta, PNS, and NMRHOLD	MOL, SVM, and BPNN	Used steering wheel status data. Using variance analysis, four parameters were selected, based on the correlation level with the driver’s status. MOL model performed best.	Accuracy: MOL: 72.92% SVM: 63.86% BPNN: 62.10%	Prepared their own dataset