THE 13th INTERNATIONAL SYMPOSIUM ON ADVANCED TOPICS IN ELECTRICAL ENGINEERING

March 23-25, 2023

Bucharest, Romania

Human Behavior, Recognition, and Interpretation

System using Physiological Signals
2023 13th International Symposium on Advanced Topics in Electrical Engineering (ATEE) | 979-8-3503-3193-6/23/$31.00 ©2023 IEEE | DOI: 10.1109/ATEE58038.2023.10108220

Ioana Raluca ADOCHIEI1,4, Oana STIRBU2, Serban Teodor NICOLESCU2, Ruxandra Victoria PARASCHIV1, Cosmin Karl
BĂNICĂ2, Felix - Constantin ADOCHIEI4
Affiliation 1: Military Technical Academy "Ferdinand I", [email protected]
Affiliation 2: SC WING COMPUTER GROUP SRL, [email protected], [email protected], [email protected]
Affiliation 3: Titu Maiorescu University, [email protected]
Affiliation 4: DMAECS, University Politehnica of Bucharest, Romania,
Abstract: Over the last decade, the study of human- factors can also affect the amplitude of the EEG signals,
computer interaction has significantly expanded in so it is only sometimes a direct measure of the number of
response to the population's needs, particularly in neurons involved. Usually, the amplitude of the waves is
understanding human behavior. Much of this area's around 100 µV. Figure 1 illustrates a view of the 19 EEG
research exploits emotions from facial and speech
signals’ evolution in time. Depending on the recorded
information. However, emotions are not always manifested
through facial and vocal communication. Instead, emotions
frequencies, five from low to high-frequency bands are
are closely related to behavioral expression, which can be distinguished: alpha (α), theta (θ), beta (β), delta (δ), and
consciously controlled and modified, leading to subjective gamma (γ). Alpha waves occur in the back half of the
interpretation. Therefore, other approaches to emotion head and can be observed in all areas of the brain’s posterior
recognition have emerged, focusing on physiological lobes [2-3]. Our main objective was to determine the type
information such as heart rate, brain electrical activity, of human temperament. We considered alpha waves
galvanic skin response, a.s.o. This article introduces a more attractive in this study, given their association with
system that detects and interprets behavioral patterns distinct kinds of human character, as reported in the
when a human subject is exposed to external stimuli scientific literature [4]. The following presents the results
through electroencephalographic signals.
of determining temperament type using the test method
Keywords: Convolutional Neural Network, alpha waves, and EEG wave processing.
temperament test
II. MATERIALS AND METHODS
I. INTRODUCTION
One of the research’s objectives was to determine the
The human brain's neural activity begins between the type of temperament using Convolutional Neural
17th and 23rd week of prenatal development. It is Network (CNN). As part of the research, we conducted
believed that from this early stage and throughout life, psychophysiological testing (temperament tests) on 100
the electrical signals generated by the brain characterize participants. The results of the temperament tests were
the functioning of the brain and the state of the whole analyzed by a psychologist and considered as the study’s
body [1]. Electroencephalography (EEG) is a non- reference. We also processed EEG waves by determining
invasive method that graphically records the variation of the exact characteristics of the subjects' signals. The
the potential difference between two electrodes placed in recorded calls underwent three stages: preprocessing,
the vicinity of the brain, at least one of which records the feature extraction, and classification.
electrical activity of the brain.
III. INSTRUMENTS AND MATERIALS

A high-resolution multichannel portable EEG

equipment with 16 electrodes (14 channels with two
reference channels) [5] was used for EEG signal
acquisition, Figure 2. The system comprises several
components, including scalp electrodes for signal
acquisition, a signal isolation amplifier for enhanced
signal quality, an analog-to-digital converter for
digitizing the acquired signals, and a wireless
Fig. 1. Representation of EEG waves evolution in time [1] transmission module for transmitting the digitized
The amplitude of the recorded waves can indicate signals. The electrodes were placed at specific locations
synchronous neuronal activity at a given time. Still, other on the scalp, following the international 10-20 system,

ISBN: 979-8-3503-3193-6/23/$31.00 ©2023 IEEE

Authorized licensed use limited to: Polytechnic University of Bucharest. Downloaded on January 20,2025 at 11:41:18 UTC from IEEE Xplore. Restrictions apply.
which included positions at F3, FC5, AF3, F7, T7, P7, independent variables were the temperament test
O1, O2, P8, T8, F8, AF4, FC6, and F4. responses and each subject's database of acquired EEG
waves. The dependent variables are temperamental
characteristics: emotionality, activity, and extraversion.
Again, these characteristics result from processing test
responses and EEG waves.
This research focused on alpha waves associated with
a frequency between 8 and 13 Hz. We noted that in the
scientific literature, individual alpha wave frequency
stands out as a measure that can be correlated with
personality dimensions. In particular, some studies have
Fig. 2. EMOTIV EPOC X headset reported a link between alpha wave frequency and four
temperament types, as shown in Table II [9-10].
The algorithms for the EEG wave processing were
developed in Matlab R2022b. The preprocessing of the TABLE II
raw data involved removing components caused by noise RELATIONSHIP BETWEEN TEMPERAMENT TYPES AND ALPHA
sources and artifacts. After preprocessing, the data was WAVES
used to construct time and frequency representations. Temperament Mean alpha wave (Hz) Alpha wave range
Then, we extracted the 11th standard EEG waves, which
characterize the brain’s neural activity from the signals’ Sangvinic 8 7,95 – 8,35
power spectral density (PSD). The signal's power Phlegmatic 9,5 9,35 – 9,85
spectral density, the base for calculating the frequency Choleric 10,5 10,15 – 10,65
domain features, can be calculated with several Melancholic 12 11,46 – 12,25
parametric or non-parametric methods. Non-parametric
methods are used more often and include techniques like
During psychological testing, we also conducted EEG
Fourier transform [6], Welch's Method [7], or
recordings to determine the type of mental disorder (if any
Thompson's Multi Taper Method [8]. In this experiment,
exist) and to be validated by licensed psychological personnel.
we employed Fast Fourier Transform (FFT) method.
This is illustrated in Figure 3.
This method uses mathematical means or tools for EEG
data analysis. The 11th standard EEG wave
characteristics and their corresponding frequency are
written in Table I.
TABLE I
BRAIN WAVES THAT WILL BE EXTRACTED

Brain wave Frequency range Mental state

Gamma 31 – 100 Hz
Increased perception
Low Gamma 31 – 40 Hz Learning
Problem-solving tasks
Middle Gamma 40 – 50 Hz
Fig. 3. EEG measurement setup
Beta 13 – 31 Hz Alert
Normal alert
Low Beta 13 – 15 Hz
consciousness To determine the four temperament types, we used
High Beta 20 – 31 Hz Active thinking CNN. First, we designed a shallow parallel CNN inspired
Alpha 8 – 13 Hz by the MobileNet technique to learn spatial
Physical and mental representation from labeled frames. Subsequently, we
Low Alpha 8 – 10 Hz
relaxation implemented a depth-wise parallel CNN using the
High Alpha 10 – 13 Hz TensorFlow framework [11].
Creativity
Perspicacity
Theta 4 – 8 Hz A. Participants
Dreams and low
awareness
Sleepiness For this experiment, we have selected 100 participants,
Delta 0.5 – 4 Hz
Dreams both genders, aged between 19 and 53 years. The
investigation started after obtaining the consent of each
Next, we considered independent and dependent participant, which required the completion of a GDPR
variables to track the differentiation of the four agreement (personal data processing agreement).
temperament types in processing EEG signals. Thus, the Participants were informed about the experiment's results

Authorized licensed use limited to: Polytechnic University of Bucharest. Downloaded on January 20,2025 at 11:41:18 UTC from IEEE Xplore. Restrictions apply.
and the outcome of the processing. The moral code and cm away from a standard LCD monitor. Activity for each
ethical constraints were respected so that the results channel was displayed on the monitor of the acquisition
would not influence the subjects' subsequent decisions computer. Preprocessing consisted of resampling each
and would not produce any negative consequences. The EEG channel at 128 Hz after applying a 50th-order FIR
identity of the issues will not be disclosed without their bandpass filter with 0.2-45 Hz cut-off frequencies. We
consent. achieved a .csv data file, which was imported into Matlab
No risks were identified for the participants during the R2022b to be processed using an algorithm based on
study. However, due to the placement of electrodes, the feature extraction.
subjects may have experienced slight discomfort. To
minimize this discomfort, short-term tests were IV. RESULTS
scheduled.
After importing the .csv files into MATLAB R2022b,
B. Procedure a conversion from the time domain to the frequency
Data were collected in the exact location, using domain was performed using an FFT. Still, removing the
identical protocols for all participants. The experiment DC offset from the data was necessary before serving this
occurred during the morning hours (10:00-11:00 a.m.). step. Figures 4 and 5 correspond to time and frequency
Therefore, participants needed to come to the representations of the processed EEG signal.
investigation rested so as not to influence the acquisition
of EEG signals or temperament test results. First, the
participants completed a temperament test: "What type
of temperament do you have?" purchased from A&C
Consulting Center [12], which involves three categories
of questions, seven types each, with two answer choices,
A or B. The three categories of questions determine the
level of emotionality (emotional and non-emotional),
activity (active and non-active), or extraversion
(extrovert or introvert). Table III presents the Fig. 4. Time representation of EEG signal
temperament types obtained by combining the six
variants.
TABLE IV
TYPES OF TEMPERAMENT

Combination Temperament
Emotional + Non-active + Extrovert Nervous
Emotional + Non-active +Introvert Sentimental
Emotional + Active + Extrovert Choleric
Emotional + Active + Introvert Passionate Fig. 5. Frequency representation of EEG signal

Non-emotional + Active + Extrovert Sanguine

Figure 6 shows the histogram of the EEG signal
Non-emotional + Active + Introvert Phlegmatic
amplitude distribution. The histogram is a visual feature
Non-emotional + Non-active + Extrovert Amorphous that illustrates the dispersion of EEG data.
Non-emotional + Non-active + Introvert Melancholic
Of these temperament categories, the interest in the
experiment was only for four temperament types:
sanguine, phlegmatic, choleric, and melancholic. Greek
physician Hippocrates (c. 460 – c. 370 BC) described the
four temperaments as part of the ancient medical concept
of humourism, that four bodily fluids affect human
personality traits and behaviors [13-14].
After completing the temperament test, the EEG Fig. 6. Histogram representation of the EEG signal
signals acquisition step followed. Participants' EEG The most superficial features of the EEG signal are
signals have been recording for 1 minute and 30 seconds. statistical features, like mean, median, variance, standard
The EEG data were acquired while the study subjects deviation, skewness, and so on [13]. Table IV shows the
numerical values of the different parameters extracted
were seated in a comfortable chair with eyes opened 70
from the raw EEG signal.

Authorized licensed use limited to: Polytechnic University of Bucharest. Downloaded on January 20,2025 at 11:41:18 UTC from IEEE Xplore. Restrictions apply.
 TABLE IV However, it is essential to note that the relationship
STATISTICAL PARAMETERS OF THE EEG SIGNAL between alpha waves and personality still needs to be
fully understood, and other factors may contribute to this
Statistical parameter Value
relationship. Additionally, the results of these studies
Minimum -1 may not apply universally and may vary depending on
Maximum 0.904671 the specific study population and methods used.
Median 0.00115164 ACKNOWLEDGMENT
Standard deviation 0.280449 This paper was developed within the project
Variance 0.0786519 "Intelligent system for evaluation and automatic
Asymmetry -0.208627
interpretation of dissimulated behavior for personnel in
critical infrastructures – VISINT", POC/222/1/3, code
Alpha waves are the focus of interest, as depicted in 122226. Financing contract: 358/390034/27.09.2021.
Figures 7-9, with time or frequency on the x-axis and
amplitude or PSD on the y-axis [14]. Butterworth REFERENCES
bandpass filter with order two has been used to filter
according to under-band frequency components: high [1] E. Niedermeyer, “The Normal EEG of the Waking Adult,”
alpha, low alpha, and alpha [15-16]. Electroencephalogr. Basic Princ. Clin. Appl. Relat. Fields, 1999.
[2] W. O. A. S. Wan Ismail et al. l “Human emotion detection via brain
waves study by using electroencephalogram (EEG),” Int. J. Adv. Sci. Eng. Inf.
Technol., 2016, doi: 10.18517/ijaseit.6.6.1072.
[3] F. Adochiei, I. Edu, et al., “Comparative Filtering Methods for Noisy
ECG Signals”, (EHB), ROMANIA, ISBN: 978-1-4577-0292-1
[3] M. A. Rahman et al. l, “Emotion recognition from EEG-based relative
power spectral topography using convolutional neural network,” Array, 2021,
doi: 10.1016/j.array.2021.100072.
[4] “Emotiv Epoc X.” https://www.emotiv.com/epoc-x/.
Fig. 7. Time representation of high alpha wave. b. Frequency representation [5] E. A. Yfantis and L. E. Borgman, “Fast fourier transforms 2-3-5,”
of high alpha wave Comput. Geosci., 1981, doi: 10.1016/0098-3004(81)90041-8.
[6] P. D. Welch, “The Use of Fast Fourier Transform for the Estimation of
Power Spectra: A Method Based on Time Averaging Over Short, Modified
Periodograms,” IEEE Trans. Audio Electroacoust., 1967, doi:
10.1109/TAU.1967.1161901.
[7] D. J. Thomson, “Spectrum Estimation and Harmonic Analysis,” Proc.
IEEE, 1982, doi: 10.1109/PROC.1982.12433.
[8] V. Trukhan et al. "Program Extension for Data Analysis from Operating
Rooms," 2020 EHB, Iasi, Romania, 2020, pp. 1-4, doi:
10.1109/EHB50910.2020.9280250
[9] I. Adochiei et al. "Complex Embedded System for Stress Quantification,"
Fig. 8. a. low alpha wave time representation of. b. low alpha wave 2019 (ATEE), Bucharest, Romania, 2019, ATEE.2019.8724892
frequency representation [10] M. Abadi et al., “TensorFlow: A system for large-scale machine
learning,”, OSDI 2016, doi: 10.48550/arxiv.1605.08695.
[11] A. C. Center, “What type of temperament do you have?”
https://www.saguna.ro/~dorulique/lectzii/2011-
2012/09D_old/09_10_26/Test-de-temperament.pdf.
[12] Adochiei F, Edu I, Schulz S, Costin H, Voss A, Changed neuronal
coupling in schizophrenia, EHB 2013, 978-1-4799-2373-1/13/$31.00
[13] D. JIANG et al., “Robust sleep stage classification with single-channel
EEG signals using multimodal decomposition and HMM-based refinement,”
Expert Syst. Appl., 2019, doi: 10.1016/j.eswa.2018.12.023.
[14] F. Adochiei, et al., "Intelligent System for Automatic Assessment and
Fig. 9. A. time representation of alpha wave. b. Frequency representation Interpretation of Disguised Behaviour for Critical Infrastructure Personnel",
EHB 2022, 2022978-1-6654-8557-9/22/$31.00 ©2022 IEEE
of alpha wave
[15] F. Adochiei, et al.," Method for Diagnosing The Level of Depression
V. CONCLUSIONS Compared with The Classical Method ", EHB 2022, Romania, November 17-
18, 2022978-1-6654-8557-9/22/$31.00 ©2022 IEEE
Our findings indicate that introverted individuals [16] T. Paraschiv, et al. l, "Method of Determining The Anxiety Level
through EEG Wave Processing ", - EHB 2022, Iasi, Romania, November 17-
exhibit high levels of dominant alpha waves, while 18, 2022978-1-6654-8557-9/22/$31.00 ©2022 IEEE
extroverted individuals tend to have lower levels. These [17] Fink, A., et al. EEG alpha power and extraversion: Individual differences
results are partially sustained by other researchers’ in affective processing. Personality and Individual Differences, 54(3), 452-
456, 2013.
outcomes [17]-[18]. Therefore, the findings related to [18] Herrmann, M. J., et al., Bilateral distribution of slow waves and alpha
introversion can be viewed as original contributions of oscillations in extroverted and introverted subjects. Personality and Individual
the authors, stemming from the results of this research. Differences, 2003.

Authorized licensed use limited to: Polytechnic University of Bucharest. Downloaded on January 20,2025 at 11:41:18 UTC from IEEE Xplore. Restrictions apply.