**1. Introduction**

Previous studies have identified that the level of driving stress could be affected by different driving conditions [1–4], such as types of roads [5,6], traffic congestion [7], and weather [8]. Dwight and David [7] identified the relationship between traffic conditions and stress levels based on driver interviews. They found that stress was higher for drivers who have experienced traffic congestions. Therefore, aggressive driving behaviors were observed more in high congestion areas than lower ones. Hill and Boyle [8] studied how different driving tasks and roadway conditions influence the stress perceived by drivers. They conducted a survey to assess drivers' stress under various roads, traffic conditions, and weather-related scenarios. The results of this study showed that driving stress was influenced by not only driver characteristics (age, gender, etc.) but also landscape types and driving distances. Therefore, we can assume that road traffic conditions and road types are associated with driving stress.

One of the most accurate indicators of driver stress is the electrodermal activity (EDA) [9–12], which characterizes the activity of electricity on human skin due to sweat [13]. Zangroniz et al. [14] found that the EDA signal is an accurate measure to distinguish calm/stressful conditions. Healey and Picard [5] presented methods for collecting and analyzing EDA data to detect driver stress during various driving conditions on actual roads. Healey and Picard found that EDA and heart rate metrics

are the most significantly correlated with driver stress. Rigas et al. [15] presented a novel methodology, based on a dynamic Bayesian network for the estimation of driver stress in specific driving events, using an electrocardiogram (ECG) and EDA signals. Singh et al. [16] studied the feature extraction method and a few algorithms for detecting stress using EDA, ECG, and photoplethysmography (PPG) signals. Munla et al. [17] used heart rate variability analysis for driving stress detection. Lal and Craig [18] studied the psychophysiological changes that occurred during a driver simulator task based on biological signals.

There has been a series of machine learning algorithms that can be applied to detect different events in various research fields by using physiological signal analysis. Plawiak et al. [19,20] applied deep genetic ensemble of classifiers to detect arrhythmia using the ECG signal and artificial neural network to estimate the state of consumption of a pump, based on dynamic pressure and vibrations. Ksiazec et al. [21] used a machine learning approach to detect Hepatocellular Carcinoma using physiological features. Rzecki et al. [22,23] proposed the computational intelligence methods for person recognition using biometric features and used the same method for the automated identification of paper-ink samples through laser-induced breakdown spectroscopy.

Also, many studies have identified that traffic conditions and road types are associated with the levels of driving stress. For the studies of the relationship between traffic conditions and driving stress, Dwight and David [24] reported that a driver's psychological state depends on the road traffic situation, and driving stress is greater in high congestion areas than in low congestion areas. Neighbors et al. [25] identified that slow traffic was linked to greater feelings of pressure and stress. Meanwhile, traffic congestion occurs when the traffic density is exceeded due to a large number of vehicles. According to the traffic flow theory [26], there are a few important traffic flow parameters, namely speed of vehicles, flow (vehicles per hour), density (number of vehicles occupying a given length of highway or lane), and road capacity. In turn, some studies [27,28] sugges<sup>t</sup> that two of the many important characteristics of road traffic are average speed and standard deviation. Table 1 shows that previous research proved the viability of vehicle speed as a characteristic of traffic congestion.



Based on available research, the current study shows a simplified hypothesis that considers only two traffic congestion parameters, namely mean speed (MS) and standard deviation of speed (STDS). The stress level of the driver was assumed to be high in high traffic conditions and low in low traffic conditions. For the studies of the relationship between road types and driving stress, Liu and Du [10] detected low, medium, and high stress levels using an EDA signal and they found that these levels corresponded to no-driving, highway driving, and city driving conditions, respectively. According to Healey and Picard [5] and Westerink et al. [6], city driving is more stressful than highway driving. Based on the previous studies, we can assume that low and high levels of stress

correspond to highway and city driving, respectively. Figure 1 shows an overview of the studied factors including traffic conditions as well as road types in this study, their impact on driving stress in terms of mental and physical, and their consequences, such as increasing risks of accidents and decreasing driving performance.

**Figure 1.** Relationship between driving stress and studied factors.

Summarizing the above information, Table 2 shows a brief compilation of studies which reflects the previous research trends in driving stress detection.




**Table 2.** *Cont.*

The distinctive features of our paper compared with previous research are following conditions and their combinations. The experimental route included city and highway roads without a driver rest time. The driver was influenced by different road types and unstable traffic conditions at the same time. EDA was used as a driving stress measure and for its analysis the special signal features were extracted. Based on Table 1 the vehicle speed signal was used as a traffic conditions indicator and new traffic congestion parameter was extracted by authors. Although many previous studies have identified the relationship between driving stress and traffic conditions, and driving stress and road types, our best knowledge has developed the classification models of driving stress by considering the traffic conditions and road types. In the previous study, most focus has been on the driver's emotional state or stress state. This study was motivated to investigate the relationship between driver stress conditions, road conditions, and road type. We tried to define and predict the traffic jam state itself considering the driver's bio-signals, in addition to the analysis of the road type. The method in this study is expected to contribute to defining a traffic jam. To do this, information was collected on actual roads for a month.
