Evaluation of Statistical Metrics by Using Physiological Data to Identify the Stress Level of Drivers Cafer Avcı 1 , Ahmet Akbaş 1 and Yusuf Yüksel 2 1 Department of Computer Engineering, Yalova, Turkey. 2 Institute of Science and Engineering, Yalova, Turkey. Abstract. In this study, an efficient and easy implementation metrics to determining stress level of drivers is proposed. For the evaluation, data have been obtained from Stress Recognition in Automobile Drivers database in the PhysioNet databank. Evaluations have been completed by using the available segment based arrays of electromyography (EMG), foot based galvanic skin response (Foot GSR), hand based galvanic skin response (Hand GSR), heart rate (HR) and instantaneous respiratory rate (IRR) derived from respiration by using peak detection algorithm. Comparisons of overall mean values of statistic metrics obtained by using EMG, Foot GSR, Hand GSR, HR and IRR showed that all of these signals and metrics can be used to identification of drivers with different physiological driving conditions. Keywords: automobile driver, stress level, electromyography, heart rate, respiration rate, galvanic skin response. 1. Introduction Biological signals express the condition of the physiological systems providing significant metrics indicating the dynamics of the internal states in human body [1]. Driver’s stress level can be measured by using the data derived from biological signals. For a driving process, data must be task related, gathered continuously and without interfering to the driver’s driving [2]. Information obtained by this way can provide a continuous measuring to specify the traffic and road conditions that affect the drivers. The information can be used to detect drivers’ stress by using in-vehicle electronic systems so that the driver's decision-making abilities can be improved [3]. Healey and Piccard conducted one of the most cited studies in this context. In this method, physiological data were collected and analyzed to evaluate driver’s stress level during real world task. The study showed that heart rate (HR) and Galvanic Skin Response (GSR) recordings are the most correlated to driver stress [4]. Ahmet Akbas concluded that Instantaneous Respiratory Rate (IRR) and average number of Contraction/Minutes (CPM) metrics can be used to determine stress level of drivers [5]. Mandeep Singh and Abdullah Bin Queyam concluded that Mean HR and Mean Hand GSR are the two statistical features related changing traffic conditions [6]. Another study studied by Mandeep Singh and Abdullah Bin, 8 features obtained from Electrocardiography (ECG), Hand GSR, Foot GSR, Electromyography (EMG) and HR recordings were classified by using Artificial Neural Network (ANN). Accuracies of this work are changed between 77.5% and 88.75% [7]. The remaining of the paper is organized as follows: obtained data for the evaluation are briefly described in section 2. Methodology, feature selection are introduced in section 3. Experimental results are summarized and described in section 4. Finally, conclusion and suggest future work that could result from paper are given in section 5.  Corresponding author. Tel.: +90 226 815 5342; fax: +90 226 815 5401. E-mail address: cafer.avci@yalova.edu.tr. 2014 3rd International Conference on Environment, Chemistry and Biology IPCBEE vol.78 (2014) © (2014) IACSIT Press, Singapore DOI: 10.7763/IPCBEE. 2014. V78. 22 124