INTRODUCTION:

The term (psychological) “stress” was coined by Hans Selye. According to him, stress is defined as “the non-specific response of the body to any demand for change”¹. Stress causes anxiety, fear and prolonged stress can lead to risk of heart-related diseases and mental illnesses such as depression. Hence, stress management is an important part of human healthcare. It is infeasible for individuals and physicians to keep track of changes in stress levels for a longer period. Hence, there is a requirement of a system which can monitor the stress levels continuously for a prolonged period of some weeks or months. Using this system, the individuals can get some alerts and be conscious regarding the change in stress levels. Physicians can use this system to monitor a patient’s mental stress levels and can analyse accordingly.

Recent advances in non-invasive wearable bio-sensors and wireless sensor networks enable it to keep track of human biological signals such as pulse rate, blood pressure, electrocardiogram (ECG), electro encephalo gram (EEG), electromyogram (EMG), galvanic skin response (GSR), etc. continuously for 24 hours a day.

A. Wireless Body Area/Sensor Network:

The applications of wireless sensor networks have considerably grown accordingly with the rapid developments in wireless and semiconductor technologies. One such application is the Wireless Body Sensor Network (WBSN) or Wireless Body Area Network (WBAN). A WBSN is a collection of low-power, miniaturized, invasive/non-invasive, lightweight wireless sensor nodes that monitor the human body functions and its environment². In WBSN, several sensors/actuators are attached to clothes and the body (non-invasive) or implanted inside the body (invasive) of a person. A sensor node is responsible for gathering data, processing them and then sending them to a control unit wirelessly. A sensor consists of following components: a power unit, a processor, memory unit and a receiver or transceiver.

A WBSN is a sensor network for a specific purpose which operates independently to connect various sensors and actuators, located in and/or on human body³. There are three important advantages of WBAN over traditional patient monitoring: (i) mobility of patients due to use of portable in/on-body monitoring devices/sensors and (ii) ability of location independent monitoring due to capability of sensors for wireless connectivity, and (iii) facility of real-time monitoring. The various sensors used in a WBAN offer different applications such as measuring the heartbeat, body temperature, blood sugar, recording ECG, EEG, EMG, GSR signals, and detecting body movements etc. Figure 1 shows a typical WBAN.

Figure 1. A typical network of Body Sensors

A WBSN can be used to monitor health of patients for long term under usual physiological conditions without coercing their common activities. WBSNs can be implemented to devise a smart and affordable healthcare system. Also, it can be a part of procedures such as diagnosis of diseases from analysis of vital signs, taking care of chronic conditions, and handling emergency events⁴.

Usually, a WBSN consists sensor networks which can be of in-body and/or on-body types. In an in-body sensor network, the communication occurs between implanted/ invasive devices/nodes and the body central unit (BCU). In case of on-body sensor network, the communication takes place between wearable/non-invasive devices/ nodes and the BCU. A WBSN sensor can connect to a network (after searching and finding a suitable one) for communication and transmit sensed data to a server (remotely located) or central control unit (CCU) for analysis and storage. It is also possible that, a WBSN can connect to the Internet autonomously and non-invasively for data transmission (Figure 2).

Figure 2. A WBAN connected to a network

Some of the challenges of WBSN is user acceptance, privacy and security of medical data. The acceptance by the patients is also an important factor for the successful execution of a new technology such as WBSN in healthcare. For that, the patients should have a clear understanding of the technology to be used for healthcare⁵. Due to huge amount of medical data collected in the due course of time, the real-time data from the sensors can be uploaded to the cloud through Internet connection, so that real-time monitoring and alert system can be realized. When the patient data is published and stored to cloud/Internet, there is always a risk of privacy and security of medical data. For this, some security measures such as maintaining secrecy⁶ and cryptographic techniques^7,8 has to be considered during implementation.

A. Galvanic Skin Response:

The skin is one of the important sensitive organ of human body which acts as a key interface between organism and the environs. The continuous variation in the electrical characteristics such as conductance of the skin is a specific property of human body, called Electrodermal activity (EDA). It reflects the integrated output of attentional, emotional and motivational processes within the sensitive nervous system acting on the human body⁹. The skin conductance also referred to as galvanic skin response is one important sensitive indicators of emotional arousal. GSR modulates the secretion of sweat from sweat glands present in the skin. Sweat glands are found in varying densities throughout the body, being the highest in the palm region (approximately 370 sweat glands per cm²) according to Gray's estimates¹⁰. The skin conductance changes in hand and foot regions due to secretion of sweat. Skin conductance is moderated autonomously by sensitive activity which initiates human behaviour, mental and emotive states on an intuitive level. GSR is one of the sensing modality for stress detection which is minimal obtrusive and has better advantage than other physiological variables because the skin is innervated, exclusively by sympathetic nervous system. Hence, GSR is highly sensitive to emotive stimulation (e.g., anger, fear, and startle response)¹¹.

B. Types of Physiological Sensors:

Researchers have designed a wide variety of wearable physiological sensors over the last two decades. One category of sensors falls into invasive or implantable type, i.e., the sensors are implanted inside the body. The other category falls into non-invasive type, i.e., these are on-body sensors. Earlier, the non-invasive type physiological sensors were wired in nature, i.e., the connection to body was through long wires, but were robust and there was no requirement of separate power supplies. But the drawback was, the user had to deal with hanging wires which was creating hindrance in mobility. A varied wireless technology for communication have been available over the last few years, from wide area networks (e.g., cellular, Internet) to short-range networks (e.g., Bluetooth¹², ZigBee¹³). Wide area networks may be used to communicate servers in a remote location (web-based healthcare¹⁴) and provide its users the freedom in mobility to a higher degree.

C. Hardware for acquisition and transmission of Physiological Sensor Data:

As discussed in previous section, the physiological sensors are capable of collecting data from human body. But for processing and analysis of those data, the sensors have to be connected to specific hardware for recording of the data. This can be achieved by a number of ways through wired or wireless medium. The hardware consists of microcontrollers¹⁵ for processing like conversion of data from analog to digital form and transceivers for transmission of data to a computer. Examples of widely used such devices are Arduino, Raspberry Pi, and so on¹⁶. For wireless environments, generally, the data can be transmitted to a remote computer using wireless sensor nodes. There are several wireless sensor nodes available by different manufacturers, such as Zolertia Z1, Coalesenses iSense, and so on¹⁷. The recorded data can be stored in a computer for processing and further analysis for real-time monitoring as well as offline applications. Figure 3 shows the hardware used for our study.

MENTAL STRESS SENSOR BASED ON GSR:

In this paper, a system has been presented which can record the GSR signals of human body in both wired as well as wireless environments and provides alerts if there is a change in stress level detected beyond a threshold.

Figure 3. Hardware used in the experimental study

A. Hardware Setup:

In this section, brief description of the hardware used or our experimental study has been provided. The hardware requirements for our experimental study are ¾ (i) a Grove GSR sensor¹⁸ for sensing GSR signal from human subject, (ii) an Arduino UNO¹⁹ board (microcontroller) used for data acquisition, and (iii) two Zolertia Z1 motes²⁰ (wireless sensor nodes having microcontroller and wireless transceivers) used for data acquisition and wireless transmission, and (iv) a computer for storing, processing and further analysis of collected data.

In our experimental study, for collecting the GSR signals from human subject, the subject has to wear the GSR sensor on the fingers. The GSR sensor is connected to the data acquisition device and the data is stored in a computer through wired or wireless medium. Figure 4 depicts the general idea of our experimental setup.

Figure 4. General experimental setup

There are two scenarios for collecting the GSR data from human body: wired and wireless. For the experimental setup in wired environment, the hardware connection has been done as shown in Figure 5. The output of the Grove GSR sensor is connected to the Arduino Uno board which is connected to a computer. The GSR sensor senses the skin conductance and sends it to the Arduino Uno board. The analog GSR signal is transformed to digital form by the Analog to Digital Converter (ADC) unit present in Arduino board, and then transmitted to the computer through the USB port for storage.

Figure 5. Hardware setup in wired environment

In case of wireless environment, two Zolertia Z1 motes are used (instead of an Arduino board) for both data acquisition and wireless transmission. One of the Z1 mote is connected to the output of GSR sensor and another Z1 mote is connected to the computer. The two Z1 motes communicate wirelessly using IEEE 802.15.4 network standard²¹. The hardware setup has been depicted in Figure 6.

Figure 6. Hardware setup in wireless environment

B. Experimental Procedure:

Experimental data from two conditions¾ mental stress and relaxation were used (not recorded due to denial by the participants for recording and storing) from volunteer participants for the experiment (all subjects provided written informed consent for the study). The participant has to sit and wear the GSR sensor in his/her two fingers of non-dominant hand as shown in Figure 7.

(a)

(b)

Figure 7. Snapshots during experiment in (a) wired environment with GSR sensor and Arduino Board (b) wireless environment with two Zolertia Z1 motes: one connected to the GSR sensor attached to hand of the subject and the other connected to a computer.

The mental stress condition imposed by conducting a sequence of tests: arithmetic operations (e.g., repeated subtraction of 9 from 500 for a duration of two minutes), memory search (a list of words to be shown in sequence for 30 seconds for memorising, then the participant has to identify them among a set of confusing words under time constraint of 1 minute), reading the sliding texts on the screen for 1 minute, Stroop colour word test²² (the participants were shown one out of five words such as violet, red, green, blue, and yellow in different colours and they had to identify the colour by clicking one of the five buttons according to the colour) for 1 minute. The relaxation condition consists of sitting calm and breathing deeply.

C. Stress Detection:

At first, during the relaxation phase (2 minutes), the mean and standard deviation of GSR values are computed. During stress conditions (conducting series of tasks) the GSR values tend to change significantly. This significant change in GSR values beyond a threshold value as compared to the precomputed mean and standard deviation of GSR values helps in stress detection.

The GSR data from the dataset²³ has been used for classification using a supervised classifier. It consists of GSR values of 10 subjects while playing computer game (a variant of PACMAN). The GSR values corresponds to frustration and normal conditions of the subjects during the play. The GSR data are preprocessed in such a way that the duration of the non-activity (values -1) were removed and divided into blocks of samples using varying window sizes of 50 to 200 depending upon the frustration conditions (0-no, 1-yes). Then the mean and standard deviation of the blocks and their corresponding frustration condition are stored. These are the features used for classification of the frustration condition (stress) of the subjects.

There are various supervised classifiers available and used for classification of different datasets^24-27. Three binary supervised classifiers: BayesNet²⁵, J48²⁶ and Random Forest²⁷ have been used for detecting the mental condition of the subject whether frustrated (stressed) or not. A 10-fold cross validation has been used for finding the accuracy of the classifications.

RESULTS:

This section presents the results obtained. Figure 8. Represents the time-series of raw GSR signals of five different subjects during the computer game playing (for a duration of 30-35 minutes). During the game playing, the subjects were mentioning the frustration (stress) conditions (0-no, 1-yes). Accordingly, the GSR data are labelled with two classes.

Figure 8. Raw GSR signals of five different subjects (subject Id S5-S9) from the dataset²³

The accuracies of various classifiers for all the 10 subjects (with blocks of window size 50) using the classifiers¾ BayesNet, J48 and Random Forest are presented in Table 1. The average accuracy of all the classifiers is shown in the Figure 9.

Table 1: Classifier Accuracy (%) for 10 subjects using BayesNet, J48 and Random Forest

Subject ID	Classifier Accuracy (%)
Subject ID	BayesNet	J48	Random Forest
S1	71.37	71.35	71.18
S2	74.48	74.66	71.88
S3	79.01	81.25	79.31
S4	85.53	85.58	84.49
S5	66.67	67.73	64.53
S6	66.2	66.46	63.85
S7	72.2	73.47	71.52
S8	71.43	72.32	66.61
S9	82.31	81.84	82.87
S10	90.66	89.98	71.89

Figure 9. Average accuracies of all the three classifiers used.

DISCUSSION:

From Table 1, it is clear that the classification accuracy of frustration detection for subjects S1 and S10 are better in case of BayesNet. Using Random Forest classifier, only it gives better result for subject S9. In case of J48 tree based classification, it outperforms for seven out of 10 subjects, i.e., S2 through S8. From Figure 9, it is clear that, J48 classifier performs better among the three with an average accuracy of about 76.5%.

In this work, a stress detector using GSR signal has been presented. Its performance analysis has also been shown. As, mental stress is also related with the emotional behaviour of a person, depending on the analysis of emotional intelligence²⁸ of a person the physician can suggest a stress management procedure to the patient. In our future work, we are trying for recording other physiological values such as heartbeat rate and EMG data from more number of subjects with their consent for study of mental emotional states of human being.

ACKNOWLEDGEMENT:

This research work is partially funded by Information Security Education and Awareness Project, Phase-II sanctioned by Ministry of Electronics and Information Technology (MeitY), Government of India.

REFERENCES:

1. What is stress? The American Institute of Stress. Available from URL: http://www.stress.org.

2. Ullah S, Higgins H, Braem B, Latre B, Blondia C, Moerman I, Saleem S, Rahman Z, Kwak KS. A comprehensive survey of wireless body area networks. Journal of medical systems. 2012 Jun 1;36(3):1065-94.

3. Khan JY and Yuce MR. Wireless body area network (WBAN) for medical applications. In ch. 31, New Developments in Biomedical Engineering, Edited by Domenico C. Signals and Communication Technology, pp. 591-627, InTech, 2010.

4. Jovanov E, Milenkovic A, Otto C, De Groen PC. A wireless body area network of intelligent motion sensors for computer assisted physical rehabilitation. Journal of NeuroEngineering and rehabilitation. 2005 Mar 1;2(1):16-23.

5. Sorathiya V, Kapupara P, Shah K. Technology Based Patient Education Information System. Research Journal of Pharmacy and Technology. 2016;9(1):101-107.

6. Anjaneyulu GSGN, Sisodiya YS, Kumar S. Novel Composite Encryption for Secrecy in Cloud Computing. Research Journal of Pharmacy and Technology. 2016; 9(9):1501-1504.

7. Vijayaragavan R. Proxy Blind Signature Scheme based on Non-Commutative Division Semi-rings. Research Journal of Pharmacy and Technology. 2016;9(7):913-917.

8. Thyagarajan K, Thangavelu A. Design and Analysis about Preventing Distributed Denial of Service Attacks in Mobile Ad-Hoc Network Using Blocking IP Broadcast Method. Research Journal of Pharmacy and Technology. 2016; 9(8):1229-1244.

9. Critchley H and Nagai Y. Electrodermal activity (EDA). In Encyclopedia of Behavioral Medicine, Edited by Gellman MD and Turner JR, pp. 666–669, Springer New York, 2013.

10. Gray H. The Organs of the Senses and the Common Integument. Anatomy of the Human Body (20th ed.). Philadelphia: Lea & Febiger. 1918.

11. Dawson ME, Schell AM, Filion DL. Handbook of psychophysiology. Cambridge University Press, Cambridge. 2000.

12. Bluetooth Technology. Available from URL: http://www.bluetooth.com.

13. ZigBee Technology. Available from URL: http://www.zigbee.org.

14. Amsa P, Pandey GK, Bhowmick M, Sonkar S, Karthikeyan D, Kumar TS. The Scope of Web-based Clinical Research in Pharmacy. Research Journal of Pharmacy and Technology. 2008;1(3):121-123.

15. Sundaresan YB, Durai MAS. An Experimental Prototype for Comparative Analysis of Intel, Atmel and PIC Microcontrollers. Research Journal of Pharmacy and Technology. 2016; 9(9):1367-1374.

16. Twelve Best Development Boards for DIY Projects. Available from URL: https:// www.mepits.com/project/236/DIY-Projects/12 -Best-Development-Boards-for-DIY-Projects.

17. List of wireless sensor nodes. Available from URL: https://en.wikipedia.org/wiki/List_of_wireless_sensor_nodes.

18. Grove GSR sensor. Available from URL: http://wiki.seeed.cc/Grove-GSR_Sensor.

19. Making Stuff with Arduino. Available from URL: http://nborko.github.io/vc-arduino.

20. Zolertia Z1, Low-power wireless module for IoT and WSN. Available from URL: http://zolertia.io/z1.

21. IEEE 802.15.4. IEEE Standard for Low-Rate Wireless Networks. Available from URL: http://standards.ieee.org/getieee802/download/802.15.4-2015.pdf.

22. Golden CJ, Freshwater SM. Stroop color and word test. Revised 2002 Adult Manual for Clinical and Experimental Uses. 2002.

23. Reuderink B, Nijholt A, and Poel M. Affective Pacman: A frustrating game for brain-computer interface experiments. In Proceedings of the 3rd International Conference on Intelligent Technologies for Interactive Entertainment (INTETAIN 2009), volume 9 of LNICST, pages 221–227, May 2009. Springer Berlin Heidelberg.

24. Chandrasekar Ravi, Khare N. Review of Fuzzy Rule Based Classification systems. Research Journal of Pharmacy and Technology. 2016; 9(8):1299-1302.

25. Friedman N, Geiger D, Goldszmidt M. Bayesian Network Classifiers. Machine Learning. 1997;29(2-3):131-63.

26. Quinlan JR. C4.5: Programs for Machine Learning. Morgan Kaufmann Publishers, 1993.

27. Breiman L. Random Forests. Machine Learning. 2001;45(1):5-32.

28. Deepak A, Krishnamoorthy K. Emotional Intelligence-A Questionnaire Based Study. Research Journal of Pharmacy and Technology. 2016;9(7):772-725.

Received on 06.12.2016 Modified on 14.12.2016

Research J. Pharm. and Tech. 2017; 10(2): 545-550.

DOI: 10.5958/0974-360X.2017.00109.3