Concealed information detection using EEG for lie recognition by ERP
P300 in response to visual stimuli: A review
MARTINA ZABCIKOVA, ZUZANA KOUDELKOVA, ROMAN JASEK
Department of Informatics and Artificial Intelligence
Tomas Bata University in Zlin, Faculty of Applied Informatics
Nad Stranemi 4511, 760 05 Zlin
CZECH REPUBLIC
Abstract: - Nowadays, lie detection based on electroencephalography (EEG) is a popular area of research.
Current lie detectors can be controlled voluntarily and have several disadvantages. EEG-based lie detectors
have become popular over polygraphs because human intentions cannot control them, are not based on
subjective interpretation, and can therefore detect lies better. This paper's main objective was to give an
overview of the scientific works on the recognition of concealed information using EEG for lie detection in
response to visual stimuli of faces, as there is no existing review in this area. These were selected publications
from the Web of Science (WoS) database published over the last five years. It was found that the Event-Related
Potential (ERP) P300 is the most often used method for this purpose. The article contains a detailed overview
of the methods used in scientific research in EEG-based lie detection using the ERP P300 component in
response to known and unknown faces.
Key-Words: - electroencephalography, EEG, lie detection, concealed information detection, EEG-based lie
detection, ERP P300, visual stimuli, known and unknown faces
Received: April 19, 2021. Revised: July 11, 2022. Accepted: August 9, 2022. Published: September 9, 2022.
1 Introduction
Recently, there has been much interest from the
scientific community in recognizing lies using various
methods. An existing device for lie detection is a
polygraph measuring the autonomic nervous system's
response. However, its accuracy and reliability vary
widely across different investigative problems.
Subjects can control their physiological responses, and
it is impossible to determine precisely whether the
subject is lying or not under stress. To overcome this
problem, brain signals are used to recognize concealed
information in the brain to detect the lie. [1] [2]
Among the frequently used techniques showing
their advantages in lie detection are
Electroencephalography (EEG), functional Magnetic
Resonance Imaging (fMRI), and functional Near-
Infrared Spectroscopy (fNIRS). [1] The most
commonly used method is EEG. [1] The EEG method
is mainly used in medicine for monitoring and
diagnosing epilepsy, stroke, seizures, or sleep
disorders. However, the EEG method has a more
extensive application, for example, in communication
and control, entertainment, or security.
The use of the EEG signal for lie detection has been
investigated since the end of the 20th century when
this area was first focused on by Farwell et al. [2].
Over the last few years, the issue of lie recognition
using EEG has developed. Researchers are devising
various methods to improve classification and high-
quality lie recognition using EEG. EEG signals can
reveal many important features of our thinking,
making it a better tool for detecting a lie. Recent
studies demonstrate the potential applicability of this
technology for lie detection. Although this idea
originated a few years ago, there are still many
opportunities for improvement, such as more powerful
classification algorithms, better availability, or lower
cost. [2]
Recent improvements in medical imaging methods
have improved our knowledge of brain function. These
techniques have enabled researchers to create
applications based on a better understanding of brain
activity. Like DNA or fingerprints, which successfully
identify the offender, another suitable option may be to
examine the offender's brain. Recent studies have
demonstrated that the brain's electrical activity can be
a reliable indicator of how information is being
processed in the brain and thus identify the
perpetrators of a crime. This method could be
beneficial and save much time in questioning
witnesses and suspects and therefore has great
potential in the criminal sciences as a new
investigative tool for linking crime evidence with
information stored in the offender's brain. Polygraphs
WSEAS TRANSACTIONS on INFORMATION SCIENCE and APPLICATIONS
DOI: 10.37394/23209.2022.19.17
Martina Zabcikova, Zuzana Koudelkova,
Roman Jasek
E-ISSN: 2224-3402
171
Volume 19, 2022
Signal
acquisition Preprocessing Feature
extraction Classification
and electroencephalographs have a significant
advantage over conventional examination methods
because they can be used in any case. [2]
It was found that the most frequently used method
in this area is Event-Related Potential (ERP) P300.
Therefore, the article's primary focus will be an
investigation using ERP P300 for EEG-based lie
detection. This research was created to provide an
overview of recent works dealing with the recognition
of concealed information for EEG-based lie detection
in the context of ERP P300 in response to known and
unknown faces, as there is no overview or summary of
current research in this area. This review will serve for
further research and identify the most successful and
frequently used methods to create an effective fraud
identification system.
2 EEG-based lie detection
2.1 Methodology
The main goal of this survey was to summarize the
most frequently used methods in studies created for
EEG-based lie detection published from January 2017
to January 2022. As far as we know, there has been no
available literature containing reviews in the field of
EEG-based lie detection focusing on visual stimuli in
the last five years. An overview of the most relevant
existing information sources in this area was compiled
to achieve the survey's main objectives. These were
specially selected publications from the Web of
Science (WoS) database according to the categories
created for this purpose. The following search query
was used for search results: (EEG OR
electroencephalogra*) AND ((lie OR decept* OR
conceal*) AND (detect* OR inform* OR decept*)).
Finally, the most relevant articles were selected to
analyze the ERP P300 component responding to
known and unknown faces for EEG-based lie detection
in the last five years.
2.2 Electroencephalography (EEG)
EEG is a noninvasive method for sensing the electrical
activity of the brain. Electrodes are placed on the
surface of the scalp. This method is most common in
medicine but can also be used in other areas such as
security, entertainment, emotion recognition, lie
detection, communication, or control. Recently, this
method has been one of the most used in the field of
lie detection.
2.3 ERP P300
Based on the type of stimuli, different types of brain
potentials are generated. One of them is ERP. It is a
subconscious psychological reaction from a reflex
generated in the human brain, measured as a result of a
motor, sensory or cognitive event in the brain while
processing information from EEG data. [3] [4] [5]
Using ERP, brain activation associated with fraud
information has been identified and is, therefore, the
primary and most widely used method for detecting
concealed information. [3] The P300 wave is an
intensively studied ERP and is its positive component.
The P300 response can be identified as a positive
deviation in the EEG signal with a typical latency of
approximately 300-1000 ms after stimulus
presentation. [7] This response is elicited in the brain
only in response to rare and meaningful stimuli in
several irrelevant stimuli generating a different
response in the subject's brain and is associated with
many processes such as attention, recognition, and
working memory. [1] [6] [7] Examining the amplitude
of the P300 wave then determines if the individual is
hiding any information. [3] [4] [5]
ERP P300 is recognized as a potent deception
detection tool because they occupy a special place due
to its most prominent peak for rare events and offers
the possibility of reliable lie detection, which is
resistant to countermeasures. [3] [8]
2.4 EEG data analysis
EEG data analysis is a complex process where each
part is essential for successful data processing and
must be solved consecutively. Fig. 1 shows a
schematic overview of EEG data analysis.
Fig. 1. EEG data analysis process.
• Signal acquisition: The first stage is signal
acquisition. EEG signals are usually recorded using
various acquisition devices such as Biosemi,
EasyCap, NeuroSky, OpenBCI, or Emotiv.
• Preprocessing: Before proceeding to data analysis,
the EEG signal must be preprocessed to remove
artifacts and noise mixed with the signal,
complicating the analysis of the stimulus-generated
ERP P300 response and reducing system
performance. [4] A bandpass filter (BPF) [4] [6] [7]
[9-14] is the most often used method to remove
noise and artifacts.
• Feature extraction: It is used to identify complex
brain wave patterns, where a useful signal is
WSEAS TRANSACTIONS on INFORMATION SCIENCE and APPLICATIONS
DOI: 10.37394/23209.2022.19.17
Martina Zabcikova, Zuzana Koudelkova,
Roman Jasek
E-ISSN: 2224-3402
172
Volume 19, 2022
selected using a set of parameters, which are then
used for classification. In previous EEG-based lie
detection studies using ERPs P300, various features
in the time, frequency, and wavelet domains have
been used to extract information, or a combination
thereof, to increase the accuracy and performance
of the system. [1] [4] [5] [7] Among the most
frequently used are Wavelet Transform (WT) [4]
[6] [10] [12], Short-Time Fourier Transform
(STFT) [14], Common Spatial Pattern (CSP) [11],
Wavelet Packet Transform (WPT) [13] and Hjorth
parameters [4] [9].
• Classification: The next step is data classification,
which is used to determine whether or not the given
information is present in the subject. The most
frequently used method in this type of research is
classification algorithms, where the resulting data
are sorted into classification classes, and the
effectiveness of the classifiers is tested. [3] [19]
The most commonly used classification algorithms
include Linear Discriminant Analysis (LDA) [4] [7]
[11] [13], Support Vector Machine (SVM) [4] [6]
[11], Multi-Layer Feed Forward Neural Network
(MLFFNN) [4] [10] [11], Naive Bayes (NB) [11],
Deep Belief Network (DBN) [12], Extreme
Learning Machine (ELM) [14] and k-Nearest
Neighbor (kNN) [9] [11].
2.5 Protocols
Nowadays, scientists use various lie identification
techniques to distinguish between guilty and innocent,
such as the Concealed Information Test (CIT) [1] [4]
[5] [6] [8 - 12] [14] [15] [18] [19], Guilty Knowledge
Test (GKT) [3] [7] and Deceit Identification Test
(DIT) [13]. These polygraphic techniques detect
psychophysiological activities, where the crime details
are known only to the guilty subject. [13] They involve
a series of questions to identify the subject's behavior.
Various studies have performed CIT, GKT, or DIT by
creating a mock criminal scenario to identify brain
potential changes in EEG's cognitive components. [4]
Compared to a polygraph, it is not so easy to deceive,
control, or suppress.
The protocols are based on recognizing a particular
stimulus, such as a murder weapon, the victim's name,
or a victim's photo. [18] The classic paradigm for these
protocols includes three categories of stimuli presented
to participants called probes, targets, and irrelevant:
• Probes: An infrequently occurring rare and
meaningful stimulus related to a crime identified
only by guilty participants. Probe images act as a
stimulus for the subject generating a P300 wave
and are images of an object or a familiar face
involved in a mock crime leading to strong memory
traces. [18]
• Targets: A non-criminal stimulus used to gain
attention and control whether the subject is
cooperating. These irrelevant items are known to all
guilty and innocent participants and generate a
P300 response. [18]
• Irrelevant: A series of irrelevant items shown to
all subjects but do not identify them as guilty or
innocent because they are unrelated to the crime
under investigation and do not generate any P300
response. [18]
It was found that the most commonly used method
for analyzing an individual's lying behavior is the CIT
method based on the ERP P300 paradigm, where the
responses to individual stimuli are examined. If P300
appears, it can be determined that the subject is lying.
This method was used, for example, by Bablani et al.
[4] [6] [9 - 12] and Dodia et al. [14] to identify fraud.
2.6 Visual ERP P300
Previous studies have further shown that faces can be
effectively used as stimuli in the context of ERP P300
to implement an efficacious lie detection system, as
the P300 component is sensitive to covert facial
recognition. Visual stimuli of known and unknown
faces based on P300 elicit different brain reactions and
thus help to identify the guilty person, e.g., whether
the subject knows the face of a particular person
(victim, accomplice, member of a terrorist group). [1]
[7] [8] [20]
2.7 The current state of EEG-based lie
detection in the context of visual ERP P300
There are a lot of research articles and scientific papers
dealing with EEG-based lie detection these days.
Many scientists have conducted different tests and
applied different approaches to the binary
classification of EEG data into guilty and innocent. [9]
The following paragraphs will summarize previous
studies on detecting concealed information for EEG-
based lie detection in the context of ERP P300 in
response to known and unknown faces.
Mehrnam et al. designed a new pattern recognition
system in response to the ERP P300 wave, which
classifies guilty and innocent subjects using the GKT
technique. The purpose was to extend the set of
properties with nonlinear elements to improve the
classification. Signals were recorded from 49 subjects.
They used BPF for preprocessing and several
morphological characteristics, frequency bands, and
wavelet coefficients for feature extraction. A genetic
algorithm (GA) was used to select the best set of
functions. They performed data analysis only on the Pz
channel. The results show that the method correctly
classified 91.83% of subjects due to combining basic
WSEAS TRANSACTIONS on INFORMATION SCIENCE and APPLICATIONS
DOI: 10.37394/23209.2022.19.17
Martina Zabcikova, Zuzana Koudelkova,
Roman Jasek
E-ISSN: 2224-3402
173
Volume 19, 2022
and nonlinear properties using the LDA classifier and
the new adaptive threshold approach. [7]
Bablani et al. proposed an approach to identifying
deception by CIT using EEG signals in ERP P300 in
response to known and unknown faces. Data collection
of 10 subjects was performed using a 16-channel
EasyCap device. Signal preprocessing was performed
by passing raw EEG signals through a BPF. This work
used Hjorth parameters (activity, mobility, and
complexity) for feature extraction and kNN as a
classifier. After performing the analysis on individual
subjects, they achieved an average accuracy of 81.9%.
[9]
Bablani et al. also used a deep learning technique
using a limited Boltzmann machine with a wavelet to
obtain information in the time and frequency domains.
They experimented on EEG data recorded by
performing CIT using a 16-channel EasyCap device by
examining the ERP P300 wave, where subjects were
presented with images of known and unknown
personalities. EEG signals were preprocessed with
BPF and analyzed by WT. To classify EEG data into
guilty and innocent people, they developed a DBN,
with an average classification accuracy of 81.03% for
10 subjects. [12]
In another study, Bablani et al. analyzed the
individual's lying behavior using the ERP P300 and
developed a new scenario for CIT. This work included
a simulated criminal scenario using a 16-channel
EasyCap device to obtain EEG from 10 subjects
recognizing the faces of known and unknown
personalities. BPF was used to remove signal-mixed
noise. They used different extraction techniques of
functions in different domains (amplitude, complexity,
mobility, frequency, power, wavelet) for a more
accurate EEG data analysis. The set framework was
developed by aggregating the results of the three best
classifiers (LDA, SVM, MLFFNN) from the five
classifiers using the classification assessment and the
weighted voting (WV) approach. The accuracy of data
classification for guilty and innocent of 84.7% was
achieved using the proposed framework (3-WV). [4]
Furthermore, Bablani et al. proposed a fraud
identification system where EEG data of 10 subjects
were obtained when performing CIT for experimental
analysis of ERP P300 in recognition of known and
unknown personalities. They used BPF for
preprocessing data of 16 channels and extracting
signals using various extraction methods. Among the
various approaches to feature extraction, WT has
proven to be the best in combination with SVM. They
proposed a new cost function where the BAT
algorithm was used to optimize SVM parameters to
increase the accuracy of the SVM classification. The
BAT binary algorithm was used to select EEG
channels. After removing non-functional canals
located in the brain's occipital lobe, the system's
performance increased to an average accuracy of up to
96.8%. [6]
In another work, Dodia et al. proposed an approach
for lie detection using EEG by performing a DIT based
on ERP P300 in response to known and unknown
faces. The experiment was performed using an EEG
acquisition device to collect data from 20 subjects. The
signals from the 16-channel EasyCap device were
preprocessed using BPF and discretized into waves
using WPT. The properties were extracted from
detailed coefficients obtained from the WPT and then
entered as input to the LDA classifier. The proposed
approach for identifying deception using WPT and
LDA resulted in a high classification accuracy of
91.67%. [13]
Further, Dodia et al. designed a CIT examining the
ERP P300, where signals obtained from 20 subjects
detected by a 16-channel EasyCap device were
preprocessed using BPF. The experiment included
reactions to pictures of celebrities and friends. Then, a
STFT method extracted features from EEG signals.
Binary BAT was used to select the optimal subset of
functions. The acquired set of features was then given
as input to the ELM classifier for training the guilty
and innocent. The resulting accuracy obtained from the
proposed lie detection system was 88.3%. [14]
In another paper, Bablani et al. proposed a hybrid
three-stage CIT classification approach that combines
the benefits of WT, k-means clustering, and MLFFNN.
The test was developed by analyzing the ERP P300
component of EEG data during a fake crime to
recognize known faces. EEG data from 10 participants
were recorded using a 16-channel EasyCap device for
CIT to implement the proposed frame and
preprocessed using BPF. The performance of the
proposed system provided an accuracy of 83.1%. [10]
In another work, Bablani et al. developed CIT
using the ERP P300 component, where subjects
observed images of known and unknown faces during
the experiment. EEG data of 7 subjects from 10
subjects were used for training and 3 for testing. BPF
was used to preprocess the EEG data obtained by the
16-channel EEG cap, and the CSP was used to feature
extraction. The fuzzy integrator system was developed
using performance indicators of classifiers as
predecessors (LDA, MLFFNN, SVM, kNN, NB).
Experimental results demonstrated an average
classification accuracy of 86.7% for three subjects
using the weighted voting approach. [11]
All these studies achieved a high classification
accuracy of about 81-97%. An overview and
comparison of particular methods for recognizing
hidden information for lie detection using EEG in the
context of ERP P300 in response to known and
unknown faces can be seen in Table 1.
WSEAS TRANSACTIONS on INFORMATION SCIENCE and APPLICATIONS
DOI: 10.37394/23209.2022.19.17
Martina Zabcikova, Zuzana Koudelkova,
Roman Jasek
E-ISSN: 2224-3402
174
Volume 19, 2022
Authors
Protocol
Dataset
Number of
subjects
EEG
device
Number of
channels
Preprocessing
Feature
extraction
Classification
Accuracy
Mehrnam et
al. (2017) [7]
GKT
Current
study
dataset
49
Ag/AgCl
electrodes
1
BPF
Combination
LDA
91.83 %
Bablani et al.
(2018) [9]
CIT
Current
study
dataset
10
EasyCap
16
BPF
Hjorth
parameters
kNN
81.9 %
Bablani et al.
(2018) [12]
CIT
Current
study
dataset
10
EasyCap
16
BPF
WT
DBN
81.03 %
Bablani et al.
(2019) [4]
CIT
Current
study
dataset
10
EasyCap
16
BPF
Combination
3-WV (LDA,
SVM, MLFFNN)
84.7%
Bablani et al.
(2019) [6]
CIT
Current
study
dataset
10
EasyCap
13
BPF
WT
SVM
96.8 %
Dodia et al.
(2019) [13]
DIT
Current
study
dataset
20
EasyCap
16
BPF
WPT
LDA
91.67 %
Dodia et al.
(2019) [14]
CIT
Current
study
dataset
20
EasyCap
16
BPF
STFT +
BBAT
ELM
88.3 %
Bablani et al.
(2020) [10]
CIT
Current
study
dataset
10
EasyCap
16
BPF
WT
MLFFNN
83.1 %
Bablani et al.
(2021) [11]
CIT
Current
study
dataset
10
EasyCap
16
BPF
CSP
Fuzzy (LDA,
MLFFNN, SVM,
kNN, NB)
86.7 %
Table 1 Comparison of existing approaches.
3 Results
For EEG-based lie detection using the ERP P300
paradigm in response to visual stimuli of known and
unknown faces, researchers in the works mentioned
above used different approaches to analyze an
individual's lying behavior. They either applied these
approaches to multiple canals [4] [6] [9] [10] [11] [12]
[13] [14] or only to the Pz canal [7].
Based on the research, it can be stated that the most
used method for analyzing the behavior of an
individual while lying is the CIT method, see
Fig. 2. Furthermore, all selected studies used their own
dataset created directly in the given articles. In Fig. 3.,
we can see that the number of subjects for the
experiment was mostly around 10. One of the most
frequently used devices for signal acquisition in
selected works is EasyCap; see Fig. 4. Fig. 5 illustrates
that most selected works focused on 16-channel data.
Furthermore, they used the BPF method for
preprocessing in all works, allowing only a specific
range of frequencies and attenuating frequency values
outside this range without reducing the signal quality.
Fig. 6 shows that the most widely used method for
feature extraction in recognition of concealed
information for EEG-based lie detection was the WT
method. The most used methods for classification were
LDA, SVM, and MLFFNN, see Fig. 7.
All of the above work used machine learning
methods and statistical approaches to data in the
brain's response to three types of stimuli: probes,
targets, and irrelevant to the detection of concealed
information stored in the brain. The purpose of the
experiments was to find out with what success rate the
method helps detect lies. The best results in the binary
classification of guilty and innocent classes in the
context of ERP P300 in response to known and
unknown faces using EEG were achieved by Bablani
et al. with an average data classification accuracy of
96.8% using WT for extraction and SVM for
classification. [6]
Another notable result is that researchers in this
area have recently focused on combining several
different methods, technologies, approaches, and
algorithms to achieve higher accuracy of EEG data
classification for lie detection. The combination of
different methods can achieve a better classification
than individual techniques. [1] [2] [4]
WSEAS TRANSACTIONS on INFORMATION SCIENCE and APPLICATIONS
DOI: 10.37394/23209.2022.19.17
Martina Zabcikova, Zuzana Koudelkova,
Roman Jasek
E-ISSN: 2224-3402
175
Volume 19, 2022
Fig. 2. The most used protocols ranged from 2017 to
2022.
Fig. 3. The most used number of subjects in given
experiments.
Fig. 4. The most frequently used techniques for signal
acquisition.
Fig. 5. The most frequently used number of channels.
Fig. 6. The most used methods for feature extraction
from 2017 to 2022.
Fig. 7. The most used methods for classification
ranged from 2017 to 2022.
CIT
78%
GKT
11%
DIT
11%
10 subjects
67%
20 subjects
22%
49 subjects
11%
EasyCap
89%
Ag/AgCl
electrodes
11%
16 channels
78%
13 channels
11%
1 channel
11%
Hjorth
11%
WT
34%
CSP
11%
WPT
11%
STFT +
BBAT
11%
Combination
22%
kNN
13%
SVM
20%
MLFFNN
20%
DBN
6%
LDA
27%
ELM
7%
NB
7%
WSEAS TRANSACTIONS on INFORMATION SCIENCE and APPLICATIONS
DOI: 10.37394/23209.2022.19.17
Martina Zabcikova, Zuzana Koudelkova,
Roman Jasek
E-ISSN: 2224-3402
176
Volume 19, 2022
4 Discussion
EEG signals can reveal many important features of our
thinking, which makes it a better tool for detecting
deception. Nowadays, scientists use the ERP P300
method to detect lies, where they examine reactions to
individual stimuli. If the P300 occurs, it can be
determined that the subject is lying. It is not as easy to
deceive, control, or suppress as the polygraph. Most of
the work has focused on examining visual stimuli by
facial recognition [4] [6-14]. There are many different
ways to visually present brain response data. One
method often effective in providing a visual
representation of differences in brain responses
involves plotting the average responses to probe,
target, and irrelevant stimuli as voltage over time at a
specific scalp location. [2] The probe stimuli are visual
stimuli such as pictures of faces, weapons, objects, or
names. However, some works have dealt with
interviews [26], audiovisual stimuli [17], name
recognition [1] [18] [22] [27], autobiographical
information [18] [21] [25], or identification of the
objects of the crime. [5] [15] [19] Different
experiments were created with mock crime scenarios
(theft [19] [23] [24]), including the victim's face [4], a
murder weapon, the accomplice's name, or a stolen
object (coin [15] [19], money, jewelry [5] [16], mobile
phone [15] [19], watch [23]). It is ascertained here
whether or not the subject participated in the given
event or is aware of the crime scene or the given
object. [2] [20]
Thanks to the development of wearable devices
containing EEG sensors, this technology is more
accessible and user-friendly. There are currently
several devices with different numbers of channels for
obtaining EEG signals, which have been used by
scientists in recent years in this field, such as EasyCap
[4] [9] [10] [12] [13] [14], Biosemi [1] [8], and Emotiv
[3]. Some researchers used only Ag/AgCl electrodes
without a headset [7].
The P300 component is often measured at the Pz,
Fz, and Cz electrodes in the skull's midline. [8] [15] In
previous studies focusing on the analysis of EEG
signals, it was found that the maximum amplitude of
this component is in the parietal lobe (Pz), the
minimum in the frontal lobe (Fz), and takes the mean
values in the central lobe (Cz). However, many
scientists have focused mainly on analyzing only one
Pz channel in the parietal area, where the amplitude of
ERP P300 is the highest. [1] [5] [7] [14] [15] [18] [19]
Most researchers have focused on lie detection
using various classification methods. However, some
have also focused on using statistical methods for
detecting lies, such as ANOVA (analysis of variance)
[28] or t-test [29]. One of the most frequently used
algorithms for classifying binary classes into guilty
and innocent (information present or absent) is LDA.
[4] [7] [11] [13]
Many authors have also worked on removing
artifacts. In order to obtain and then remove the
artifacts of blinking and eye movements in the studies,
they most often used another measurement method
such as EOG (Electrooculogram) [1] [5] [7] [8] [9]
[12] [18], which can be divided into vertical EOG
(VEOG) and horizontal EOG (HEOG). Eye artifacts
obtained by the EOG method were removed using
algorithms or visual inspection.
Another important finding is that researchers in this
field have recently focused on combining multiple
methods, technology approaches, and algorithms in
signal analysis for lie detection to achieve a higher
classification accuracy of concealed information
recognition. Some researchers have focused, for
example, on the combination of different methods such
as EEG/fNIRS [1], EEG/PPG (Photoplethysmography)
[26], and EEG/rTMS (repetitive Transcranial Magnetic
Stimulation) [24]. Some have also focused on a
combination of algorithms such as SVM, LDA,
MLFFNN, NB, and kNN [4] [11] for more accurate
fraud detection. By combining different methods,
better classification can be achieved than individual
approaches. [1] [2]
The evidence presented here and several other
studies suggest that recent developments in
neuroscience enable researchers to detect information
stored in the brain that could noninvasively,
objectively, and accurately link criminals to a specific
crime. Therefore, this method's potential is to resolve
cases faster, more accurately, and more efficiently and
provide innocent suspects with noninvasive, stress-
free, and reliable means of exemption. [2]
However, even with today's modern methods and
algorithms, 100% accuracy of lie detection has not yet
been achieved. Despite the high level of classification
accuracy that some research has achieved, there are
still several opportunities for improvement, such as
maximum classification accuracy, lower cost, better
availability, reduced time consumption, and real-time
use. Using methods for extraction, classification, and
selection of elements may be crucial, as a different
method is suitable for each type of data processing.
Emphasis is placed on the size of datasets, the type of
stimulus, or the experiment protocol when selecting
methods for extraction and classification. Because
each algorithm has a varied computing complexity and
data processing time, selecting a classifier can be
challenging. The highest success of the binary
classification of guilty and innocent data in the context
of CIT based on the examination of ERP P300 in
response to the recognition of known and unknown
faces was achieved by Bablani et al. with an accuracy
of 96.8%.
WSEAS TRANSACTIONS on INFORMATION SCIENCE and APPLICATIONS
DOI: 10.37394/23209.2022.19.17
Martina Zabcikova, Zuzana Koudelkova,
Roman Jasek
E-ISSN: 2224-3402
177
Volume 19, 2022
5 Conclusion
The central part of the article was an overview of
recent scientific research for EEG-based lie detection
using the ERP P300 paradigm in response to known
and unknown faces. The CIT method was the most
commonly used method for analyzing an individual's
lying behavior. It is evident from the survey that all
scientists used their own dataset in the selected papers,
and all used the BPF method for preprocessing. The
experiment's most common number of subjects was
around 10, and one of the most frequently used devices
for signal acquisition in selected articles is the
EasyCap. Furthermore, it turned out that most of the
selected works focused on 16-channel data. The
scientists used the WT method the most for feature
extraction in this context. The LDA, SVM, and
MLFFNN algorithms were most often used as
classifiers. Another important finding is that
researchers in this area have recently focused on
combining several methods for EEG-based lie
detection to achieve higher classification accuracy.
Recent advances in EEG mobile devices have opened
the door to many innovations in various applications.
The contribution of this study is an overview of the
most recently used methods in this area for creating an
efficient fraud detection system utilizing visual stimuli
of faces. Based on the survey, it can be concluded that
this technology has great potential for more effective
lie detection.
References:
[1] X. Lin, L. Sai, and Z. Yuan, Detecting Concealed
Information with Fused Electroencephalography
and Functional Near-infrared Spectroscopy,
Neuroscience, Vol. 386, 2018, pp. 284–294.
[2] L. A. Farwell, and S. S. Smith, Using brain
MERMER Testing to Detect Knowledge Despite
Efforts to Conceal, Journal of Forensic Sciences,
Vol. 46, 2001, pp. 135–143.
[3] S. Anwar, T. Batool, and M. Majid, Event Related
Potential (ERP) based Lie Detection using a
Wearable EEG headset, Proc. 2019 16th
International Bhurban Conference on Applied
Sciences and Technology (IBCAST), 2019, pp.
543–547.
[4] A. Bablani, D. R. Edla, D. Tripathi, and V.
Kuppili, An efficient Concealed Information Test:
EEG feature extraction and ensemble
classification for lie identification, Machine
Vision and Applications, Vol. 30, 2019, pp. 813–
832.
[5] N. Saini, S. Bhardwaj, and R. Agarwal,
Classification of EEG signals using hybrid
combination of features for lie detection, Neural
Computing and Applications, Vol. 32, 2020, pp.
3777–3787.
[6] A. Bablani, D. R. Edla, D. Tripathi, S. Dodia, and
S. Chintala, A Synergistic Concealed Information
Test With Novel Approach for EEG Channel
Selection and SVM Parameter Optimization,
IEEE Transactions on Information Forensics and
Security, Vol. 14, 2019, pp. 3057–3068.
[7] A. H. Mehrnam, A. M. Nasrabadi, M. Ghodousi,
A. Mohammadian, and S. Torabi, A new
approach to analyze data from EEG-based
concealed face recognition system, International
journal of psychophysiology, Vol. 116, 2017, pp.
1–8.
[8] A. Alsufyani, O. Hajilou, A. Zoumpoulaki, M.
Filetti, H. Alsufyani, Ch. J. Solomon, S. J.
Gibson, R. Alroobaea, and H. Bowman,
Breakthrough Percepts of Famous Faces,
Psychophysiology, Vol. 56, 2019.
[9] A. Bablani, D. R. Edla, and S. Dodia,
Classification of EEG Data using k-Nearest
Neighbor approach for Concealed Information
Test, Procedia Computer Science, Vol. 143,
2018, pp. 242–249.
[10] A. Bablani, D. R. Edla, V. Kuppili, and D.
Ramesh, A Multi Stage EEG data classification
using k-Means and Feed Forward Neural
Network, Clinical Epidemiology and Global
Health, Vol. 8, 2020, pp. 718–724.
[11] A. Bablani, D. R. Edla, V. Kuppili, and R.
Dharavath, Lie Detection Using Fuzzy Ensemble
Approach With Novel Defuzzification Method for
Classification of EEG Signals, IEEE Transactions
on Instrumentation and Measurement, Vol. 70,
2509413, 2021, pp. 1–13.
[12] A. Bablani, D. R. Edla, and V. Kuppili, Deceit
Identification Test on EEG Data Using Deep
Belief Network, 2018 9th International
Conference on Computing, Communication and
Networking Technologies (ICCCNT), 2018, pp.
1–6.
[13] S. Dodia, D. R. Edla, A. Bablani, D. Ramesh, and
V. Kuppili, An Efficient EEG based Deceit
Identification Test using Wavelet Packet
Transform and Linear Discriminant Analysis,
Journal of Neuroscience Methods, Vol. 314,
2019, pp. 31–40.
[14] S. Dodia, D. R. Edla, A. Bablani, and R.
Cheruku, Lie detection using extreme learning
machine: A concealed information test based on
short-time Fourier transform and binary bat
optimization using a novel fitness function,
Computational Intelligence, Vol. 36, 2019, pp.
637–658.
[15] A. Akhavan, and M. H. Moradi, Detection of
Concealed Information Using Multichannel
WSEAS TRANSACTIONS on INFORMATION SCIENCE and APPLICATIONS
DOI: 10.37394/23209.2022.19.17
Martina Zabcikova, Zuzana Koudelkova,
Roman Jasek
E-ISSN: 2224-3402
178
Volume 19, 2022
Discriminative Dictionary and Spatial Filter
Learning, IEEE Transactions on Information
Forensics and Security, Vol. 13, 2018, pp. 2616–
2627.
[16] J. Gao, J. Song, Y. Yang, S. Yao, J. Guan, H. Si,
H. Zhou, S. Ge, and P. Lin, Deception Decreases
Brain Complexity, IEEE Journal of Biomedical
and Health Informatics, Vol. 23, No. 1, 2019, pp.
164–174.
[17] W. Chang, H. Wang, Z. Lu, and Ch. Liu, A
Concealed Information Test System Based on
Functional Brain Connectivity and Signal Entropy
of Audio-Visual ERP, IEEE Transactions on
Cognitive and Developmental Systems, Vol. 12,
No. 2, 2020, pp. 361–370.
[18] G. Lukacs, A. Grzadziel, M. Kempkes, and U.
Ansorge, Item Roles Explored in a Modified
P300-Based CTP Concealed Information
Test, Applied Psychophysiology and Biofeedback,
Vol. 44, 2019, pp. 195–209.
[19] A. Akhavan, M. H. Moradi, and S. R. Vand.
Subject-based discriminative sparse
representation model for detection of concealed
information, Computer Methods and Programs in
Biomedicine, Vol. 143, 2017, pp. 25–33.
[20] L. A. Farwell, Brain fingerprinting: a
comprehensive tutorial review of detection of
concealed information with event-related brain
potentials, Cognitive Neurodynamics,
Vol. 6, 2012, pp. 115–154.
[21] M. Funicelli, L. White, S. Ungureanu, and J. R.
Laurence, An Independent Validation of the
EEG-Based Complex Trial Protocol with
Autobiographical Data and Corroboration of its
Resistance to a Cognitively Charged
Countermeasure, Applied Psychophysiology and
Biofeedback, Vol. 46, 2021, pp. 287–299.
[22] A. Alsufyani, K. Harris, A. Zoumpoulaki, M.
Filetti, and H. Bowman, Breakthrough percepts of
famous names, Cortex, Vol. 139, 2021, pp. 267–
281.
[23] P. Liu, H. Shen, and S. Ji, Functional
Connectivity Pattern Analysis Underlying Neural
Oscillation Synchronization during Deception,
Neural Plasticity, Vol. 2019, 2019, p. 10.
[24] I. Karton, and T. Bachmann, Disrupting
dorsolateral prefrontal cortex by rTMS reduces
the P300 based marker of deception. Brain
Behavior, Vol. 7, No. 4, 2017.
[25] Q. Liu, X.G. Zhao, Z.G. Hou, and H.G. Liu, Deep
Belief Networks for EEG-Based Concealed
Information Test, Advances in Neural Networks,
Vol. 10262, 2017, pp. 498–506.
[26] M. D. Kohan, A. M. Nasrabadi, M. B.
Shamsollahi, and A. Sharifi, EEG/PPG effective
connectivity fusion for analyzing deception in
interview, Signal Image and Video Processing,
Vol. 14, 2020, pp. 907–914.
[27] S. Thakur, R. Dharavath, and D. R. Edla, Spark
and Rule-KNN based scalable machine learning
framework for EEG deceit identification,
Biomedical Signal Processing and Control, Vol.
58, 2020.
[28] Y.F. Lai, M.Y. Chen, and H.S. Chiang,
Constructing the lie detection system with fuzzy
reasoning approach, Granular Computing, Vol. 3,
2018, pp. 169–176.
[29] W. Chang, H. Wang, Ch. Hua, Qi. Wang, and Y.
Yuan, Comparison of different functional
connectives based on EEG during concealed
information test, Biomedical Signal Processing
and Control, Vol. 49, 2019, pp. 149–159.
Contribution of individual authors to
the creation of a scientific article
(ghostwriting policy)
Martina Zabcikova was responsible for the overall
research progress and writing the paper.
Zuzana Koudelkova participated in the survey,
concept, and verification of the results.
Roman Jasek was responsible for the supervision and
conceptualization of the article.
Sources of funding for research
presented in a scientific article or
scientific article itself
This work was supported by IGA (Internal Grant
Agency) of Tomas Bata University in Zlin under the
project No. IGA/CebiaTech/2022/006.
Creative Commons Attribution License
4.0 (Attribution 4.0 International , CC
BY 4.0)
This article is published under the terms of the
Creative Commons Attribution License 4.0
https://creativecommons.org/licenses/by/4.0/deed.en_
US
WSEAS TRANSACTIONS on INFORMATION SCIENCE and APPLICATIONS
DOI: 10.37394/23209.2022.19.17
Martina Zabcikova, Zuzana Koudelkova,
Roman Jasek
E-ISSN: 2224-3402
179
Volume 19, 2022
