Driving Anger States Detection Based on Incremental Association Markov Blanket and Least Square Support Vector Machine

Abstract

Driving anger, known as “road rage”, has gradually become a serious traffic psychology issue. Although driving anger identification is solved in some studies, there is still a gap in driving anger grading which is helpful to take different intervening measures for different anger intensity, especially in real traffic environment. The main objectives of this study are: (1) explore a novel driving anger induction method based on various elicitation events, e.g., traffic congestion, vehicles weaving/cutting in line, jaywalking and red light waiting in real traffic environment; (2) apply incremental association Markov blanket (IAMB) algorithm to select typical features related to driving anger states; (3) employ least square support vector machine (LSSVM) to identify different driving anger states based on the selected features. Thirty private car drivers were enrolled to perform field experiments on a busy route selected in Wuhan, China, where drivers’ anger could be induced by the elicitation events within limited time. Meanwhile, three types of data sets including driver physiology, driving behaviors and vehicle motions, were collected by multiple sensors. The results indicate that 13 selected features including skin conductance, relative energy spectrum of β band of electroencephalogram, standard deviation (SD) of pedaling speed of gas pedal, SD of steering wheel angle rate, vehicle speed, SD of speed, SD of forward acceleration and SD of lateral acceleration have significant impact on driving anger states. The IAMB-LSSVM model achieves an accuracy with 82.20% which is 2.03%, 3.15%, 4.34%, 7.84% and 8.36% higher than IAMB using C4.5, NBC, SVM, KNN and BPNN, respectively. The results are beneficial to design driving anger detecting or intervening devices in intelligent human-machine systems.

A Generic Design of Driver Drowsiness and Stress Recognition Using MOGA Optimized Deep MKL-SVM

“…Electrodermal activity (EDA) skin potential response (SPR) is another typical input for the driver stress detection, which an example was shown along with adaptive filtering and spike detection techniques [29]. Skin conductance and EEG served as inputs of the driver stress condition, which were further explored by the incremental association Markov blanket algorithm for feature extraction [30]. The least square support vector machine (LS-SVM) was applied to construct the classifier.…”

“…Various methods have been proposed for driver drowsiness recognition [16][17][18][19][20][21][22][23][24][25][26][27] and driver stress recognition [28][29][30][31][32][33][34][35]. Nonetheless, there exist several challenges and limitations that require further research.…”

“…Consider driver stress recognition, Table 3 compares the performance between the proposed algorithm and related works [28][29][30][31][32][33][34][35], towards a category of input signals, dataset, methodology, types of cross-validation and performance. Similar to driver drowsiness recognition, related works in driver stress recognition experienced similar issues.…”

“…The EEG signal was adopted in [30], which the input signal has a measurement stability of 85% [37]. Some of the works [28,29,33] did not analyze the method using cross-validation.…”

“…It is noted that there is an unspecified dataset in [35]. To compare the rest of the existing works, the proposed algorithm outperforms [30][31][32]34] significantly.…”