Computational Intelligence and Soft Computing Paradigm for Cheating Detection in Online Examinations

. Covid-19 has been a life-changer in the sphere of online education. With complete lockdown in various countries, there has been a tumultuous increase in the need for providing online education, and hence, it has become mandatory for examiners to ensure that a fair methodology is followed for evaluation, and academic integrity is met. A plethora of literature is available related to methods to mitigate cheating during online examinations. A systematic literature review (SLR) has been followed in our article which aims at introducing the research gap in terms of the usage of soft computing techniques to combat cheating during online examinations. We have also presented state-of-the-art methods followed, which are capable of mitigating online cheating, namely, face recognition, face expression recognition, head posture analysis, eye gaze tracking, network data trafc analysis, and detection of IP spoofng. A discussion on improvement of existing online cheating detection systems has also been presented.


Introduction
Online learning has been in demand for the past couple of years.However, the COVID-19 pandemic, since late 2019 to early 2020, has thrusted the need for online learning in a very signifcant manner.With schools and higher education institutions which closed down around the world, there was a dire need for classroom-like engagement of around 1 billion students, to provide continuous learning.Engaging online classes without much practical exposure for students was a huge challenge.However, a bigger challenge has been to evaluate the performance of such students through online examinations.Fakhroddin et al. [1] have performed a systematic review on the various behaviours of cheating during online examinations.Tey provide a comprehensive view on methods of cheating detection, prevention, and mitigation.Kaddoura et al. [2] have presented a study on the steep rise of online learning portals which indirectly poses the challenge of fair conduction of online examination.In comparison to an ofine examination, it is easier to cheat during an online examination.
Monitoring of online examinations through human proctors is a common methodology to prevent cheating.However, the con of this methodology is the cost borne to employ individuals to monitor the examinations taken.Tere is a requirement of high bandwidth for communication in such cases and there further exists no metric to evaluate the efciency of the proctor in cheating detection.Semiautomated proctoring has also been proposed in several research studies such as the ones presented by presented by Rosen and Carr [3] and Li et al. [4], where a desktop robot transmits videos to a proctor for monitoring any suspicious motion in the video.Proctoring through authentication, keystroke recognition, and through webcams ofer a level of security in online exams (Xiong and Suen) [5].Various researchers have performed online proctoring using webcams for face recognition and text detection [6,7].Many research studies have also been conducted to monitor the variation of head posture of students to detect possible cheating [8,9].Time-to-time authentication and eye tracking have been adopted in [10,11] to ensure detection of online cheating.Te advances in computer networking have also been leveraged to detect cheating through tracing malicious data packets or data sent to blacklisted URLs [12,13].
However, these exists a lot of other literature to improve techniques for the detection of cheating in online examination.Te motivation behind writing this research article is to analyse what are the categories of methods used for the detection of online cheating using soft computing.Furthermore, we explore the potential techniques that could be employed for online cheating detection which have not been used yet for the same.Te main contribution of our article is towards presenting a comprehensive literature review of state-of-the-art soft computing techniques available for the detection of cheating during online examinations.We also present possible solutions using varied methods to mitigate cheating during online examinations.Te scope of the article is to present the various methods available for the detection of cheating in online examinations.Te next section presents the research gap and methodology followed to address the research gap.

Systematic Literature Review for Identifying the Research Gap and Methodology
2.1.Process of Systematic Literature Review.SLR is a method that makes it possible to gather pertinent data about a specifc subject that meets the preestablished eligibility requirements and provides a solution to the research questions that had been posed [14].Our review article presents the existing techniques available for the detection of cheating in online examinations and further proposes the usage of improved techniques with better efciency for the same task.Figure 1 presents the SLR and meta-analysis diagram which has been followed in our literature review for the process of screening and evaluation.Table 1 further elaborates on the steps followed, outcomes, and methods adopted at every level of the SLR process.

Research Gap. With the advent of online learning during
Covid-19, a need has arisen to monitor cheating behaviour during online examination.In spite of considerable research related to the modes of online learning, there is no comprehensive review of the techniques available or which could be adopted to mitigate cheating during online exams.Our research article answers the following research questions: (i) What are the categories of methods used for detection of online cheating using soft computing?
(ii) What are the potential techniques that could be employed for online cheating detection which have not been used yet for the same?
Terefore, our research article addresses this research gap and satisfes the following objectives pertaining to detection of cheating during online examinations: (i) To present a comprehensive literature review of state-of-the-art soft computing techniques available for the detection of cheating during online examinations (ii) To present possible solutions using varied methods to mitigate cheating during online examinations.

Methodology.
Table 1 presents the overall methodology of selection and categorization of the articles for our review.Figure 2 showcases the broad categorization of methods followed for the detection of online cheating.Tey are broadly classifed under the categories of the following subsections:

Face Tracking.
Tis technique usually employs a camera to detect the face of the subject taking the online examination.By monitoring his gestures, researchers can develop systems to alert proctors during online cheating.

Face Expression Detection.
On the same lines of face tracking, the face expression detection technique is used to monitor stressful or sly expressions on the face of the student while taking an online exam and this can be used to alert the proctor.

Head Posture Analysis.
Head posture analysis detects if the subject taking the exam bends his head in a certain direction beyond a threshold value which could suggest that he is involving in cheating.

Eye Gaze
Tracking.Eye gaze tracking systems are used to monitor the alertness of drivers and pilots while driving and fying aircrafts, respectively.Te same mechanism is employed to detect if a person is fxing his gaze in a particular direction for an unusually long time.It can also detect the chances of a person passing sly glances to his neighbour or surrounding persons during an online exam.

Network Data Analysis and Trafc Classifcation.
While taking an online exam, the student is free to access any resource over the internet.In spite of blocking of essential websites which may lead to copying, there is still a possibility of the student accessing the internet in search for answers.

Applied Computational Intelligence and Soft Computing
Te technique of network data trafc analysis and classifcation can be used to detect the type and content of trafc emanating from a student's system while taking the examination online.
2.3.6.IP Spoofng Detection.IP spoofng is the mechanism of one computer system masquerading as another.Tere could be cases where the genuine student registered to take the online exam would allow a proxy user to take the exam in his place.Tis can be detected via advanced online IP spoofng detection mechanisms.Tis article has been divided into the following sections: Section 3 deals with face detection and facial expression detection systems.Section 4 highlights the techniques used for the detection of change in head posture.Section 5 of this article lays emphasis on the literature available for eye gaze tracking.Section 6 of the article presents the state-of-the-art techniques for network data trafc analysis and classifcation.Section 7 presents the mechanisms used for the detection of IP spoofng.Section 8 presents a brief discussion of the usage of an ensemble of all the techniques to model a robust online cheating detection mechanism to aid authentic online learning.

Existing Techniques for Detection of Cheating Using Face
Tracking and Facial Expression Detection.Image processing techniques have been used for tracking faces and facial expressions to classify and predict behaviours in people.Tis section of the article describes some of the literature that uses face tracking and facial expression detection algorithms to detect cheating during online examinations.Figure 3 adapted from Kaddoura and Gumaei [6] showcases the usage of a webcam and backcam to detect malicious activity by the student.Te webcam provides a front view of the face of the student, and the data retrieved from this can be used to monitor his expressions, whereas the backcam gives view of the material he is accessing (second column of images), which in this case is a text book and a cheat sheet.
Fan et al. [7] presented the usage of Microsoft Kinect, a webcam device for gesture detection, by measuring head roll and yaw angles.Under the table, hand movement can also be detected using this device.Te classifcation of images is performed by thresholding.Any roll or yaw movement beyond a certain threshold is treated as malicious activity and referred to as cheating.Unfortunately, this methodology may prove to be less efective when considering a large population of students, each of whom would have diferent postures of seating.Many students may have the tendency to lean forward or sideways while writing, and this could get misclassifed as cheating.Te benefts of high accuracy have been leveraged in [15] where a CNN coupled with a Haar Cascade classifer and Viola-Jones feature extractor has been used for the detection of cheating in students.
He et al. [16] have presented a face tracking system with a database of faces of all students writing the exam.Tey claim to detect "ghost writers," i.e., students/individuals whose faces are not registered in the system.Tis system typically prevents a proxy writer from taking an online exam on behalf of another student.Tough the system may prove to be efcient for small numbers, as the number of students get scaled up, the system may fail to perform.Furthermore, a large dataset of all faces of individuals taking the exam would be required which leads to a huge storage capacity requirement.
Most techniques discussed in Section 3.1 are rudimentary and pose a scope of improvement.Section 3.2 outlines novel techniques used for face tracking and the detection of facial expressions which could be potentially adapted to design systems for the detection of online cheating.

Potential Techniques for Face Tracking and Expression
Detection for Online Examinations.Articles from the literature by Ren et al. [17], Liu et al. [18], and Yuankai et al. [19] have explored the usage of deep neural networks for the extraction of facial features for image recognition (Table 2).For instance, Ren et al. [17] have used the Chinese whisper algorithm for extraction of facial features and double triplet NN for classifcation of the images.Gan and C. P. [18] have been able to overcome the problem of a person writing an exam in various poses by developing a pose invariant face recognition (PIFR) algorithm.Tis would be of great help for students who have diverse postures while writing the exam as various profles of their faces would be registered and facial feature matching would be done.In Jianwen et al.' study [20], a globally optimized modular boosted ferns (GoMBF)-cascade regression model has been constructed to prove its superiority in feature detection when compared to the classical explicit shape regression (ESR) algorithm (Figure 4).
Apart from general face detection, soft computing algorithms have also been used for the detection of various facial expressions.Tere are various biomarkers/biological indicators of stress, panic, grief, joy, and others that are displayed by a human body.When these markers are detected, tagged, and classifed appropriately, they could prove     3 presents a comparison of the latest literature, apart from conventional image processing techniques, for detection of facial expressions.Mohammad et al. [22] have devised a novel technique for the detection of facial expressions using EEG signals along with the long-short-term memory recurrent neural network (LSTM-RNN) for the purpose of classifcation. Figure 5 illustrates the power spectral density of the alpha, beta, gamma, and theta waveforms of a human subject while experiencing varied emotional states of fear, anger, calm etc.For the purpose of online examinations, this technique could prove to be helpful as whenever a student would have thoughts of cheating or involve in the act of cheating, the behaviour could be extracted from the PSDs of the EEG signals.Tese data could be used to supplement the image data being gathered by a webcam and fed to a neural network for accurate facial expression recognition.

Applied Computational Intelligence and Soft Computing
Xinyu et al. [23] used yet another biomarker of oxygen saturation levels in the facial tissues during stressful situations.Hyperspectral imaging is used to detect these stress levels and can be appropriately mapped to classes of behaviours.Tis could prove to be of great help for cheating detection.On similar lines of [23], Lanxin et al. [24] used the linear discriminant analysis (LDA) for the classifcation of facial expressions.
Apart from detection over a local system, the transference of data through the internet to remotely placed proctors is also a key area of research.Jiannnan et al. [25] described a novel technique of detection of facial expressions using facial action unit (AU), which is a unit assigned to the expressions based on the change of muscles of the face while feeling the assigned emotion.Te authors have also presented a light-weight method (using Raspberry Pi) to transfer these data from the client (student) system to any remote server (proctor) by leveraging the low latency Applied Computational Intelligence and Soft Computing provided by edge computing.Tis methodology would be a boon for our application of cheating detection, considering that several real-time facial expressions would have to be classifed and transferred to the proctor in real time for him to take further action.Tis section presented the state-of-the-art and novel image processing and soft computing techniques for face tracking and facial expression detection.As discussed, several of these techniques can be employed for better accuracy of detection of cheating, lower false positive rates, and most importantly, less network transmission overhead.Section 4 of this article presents various techniques for the detection of cheating in online examinations using head posture analysis.

Head Posture Analysis Systems
Tere are several studies which have been performed to monitor the posture of a person's head while he sits before a computer/laptop.Tese studies have helped in detection of potential cheating activity by noticing the change in the roll, pitch, and yaw angle of the head (Figure 6) beyond a certain  When the threshold gets crossed, the activity of the person may be deemed as malicious and suitable methods could be applied to tackle the same.Hu et al. [26] used a method where a webcam is attached to the monitor of the computer or laptop on which the student is expected to take the online examination.From the same, an image of the person is input which is passed through the Adaboost algorithm with the Haar classifer to detect the movement of the head beyond a certain threshold degree.Once the threshold is crossed, a malicious activity is detected.A similar principle of thresholding using various other classifcation algorithms using multivalued variables has been proposed by Prathish and Bijlani [27], Zhu and Ramanan [28], and Geng and Xia [29].
Donghoon et al. [30] have leveraged the benefts of convolutional neural networks along with the random forest algorithm to perform the classifcation of head postures beyond a posture.Due to the dense layers of features extracted using the CNN model, the authors have been able to achieve accuracies of the order beyond 97%.
Yu et al. [31] presented interesting results for creating a 3D morphable model using ResNet.Te novelty of this system stems from the fact that not just is a thresholding mechanism used for head pose analysis, but even a 3D model which changes according to the change in head posture of the student, has been created.Tis could be a revolutionary technique for monitoring students during an online examination.Hu et al. [32], Bouaziz et al. [33], and Taigman et al. [34] have presented their results of 3D modelling of faces on the same lines as Yu et al. [31], which can be explored for their usage in cheating detection during online exams.
Rajeev et al. [35] have presented the usage of ResNet for classifcation of facial features, head poses, and gender of the student using the CelebA dataset (shown in Figure 7).Similar approaches with varied accuracies have been proposed and proven by Hu et al. [32], Bouaziz et al. [33], and Taigman et al. [34].
Hobeom et al. [36] have presented the development of magnet-magnetometer pairs for monitoring the change in head posture.Based on the craniovertebral angle of the person, a Gaussian regression model is generated to estimate further changes in CVA.Based on these changes, the decision tree and support vector machine classifcation algorithms have been explored to classify the changing head postures.
Various other methods used for head posture analysis have been described in Table 4.
Terefore, this section presented the possibilities of usage of various head posture analysis techniques to combat the problem of cheating detection in online examinations.Section 5 of this article presents a review of various eye gaze tracking systems that are available in the literature.Te usage of the eye gaze tracking systems also proves as one of the methods for online exam cheating detection.

Eye Gaze Tracking Systems
Te technique of eye tracking computes the eye gaze point of a student as he looks at his surroundings.Based on where he is looking, the rectangular coordinates of the form (x, y) are computed for the subject with the screen as a reference point [42].Several applications of this technology have been explored, namely, active and passive applications.Active   applications include control of devices using eye gaze, login using eye activation, etc. Passive applications include analysis of design, vehicular safety, medical diagnosis, etc. [43].Based on Bawarith et al. [10], eye gaze tracking with the usage of logistic regression has been utilized for the detection of malicious activity by a subject during an online examination.Similar approaches have been followed by Curran et al. [44], Kerkvliet and Sigmund [45], Keresztury and Cser [46], and Yee and MacKown [47].
Complex eye gaze tracking systems have been developed by researchers for various other purposes, and the same can be utilized for the detection of cheating during online examinations.
Johannes et al. [21] have proposed the development of an eye gaze tracking sensor coupled with an augmented reality system (Figure 8).A gaze angle resolution of 2.3 °has been achieved using this system.Terefore, any minute change in the eye gaze can be detected using this system.Crossing the eye gaze threshold by a large margin may be deemed as a malpractice in the case of online examinations by utilizing such as system.
Bazrafkan et al. [48], Braunagel et al. [49], Corcoran et al. [50], and Bissoli et al. [51] have also presented the development of eye gaze tracking systems for estimation of drivers' attention while driving, consumer electronic applications, development of HMI systems, etc. Table 5 reviews diferent state-of-the art systems available for eye gaze tracking which may be useful for cheating detection in online examinations.
Apart from the aforementioned techniques for tracking cheating in online exams, one of the major methods followed for the same could be classifcation of computer network trafc arising from the subject's computer.Section 6 presents various methodologies applied for detection and classifcation of computer network trafc which could serve as a major breakthrough for cheating detection.

Network Data Analysis and Traffic Classification
During an online examination, the student has free internet access to various sources of information and related study material.Te detection of the content that he is browsing on the internet while taking the exam is a critical technique to classify his behaviour as malicious or not.A simple method to check the content of his screen would be to mount a camera behind the subject to monitor his screen all the time.However, such a system would create a lot of overhead in terms of data captured and relayed to the proctor who would also be monitoring the face of the student.Terefore, the need arises to develop a mechanism to monitor the browsing activity of the student and analyse the data trafc generated from his system while he takes the exam.Tis would include the websites he tries to visit, the amount of time those websites are open, and any content that he might have downloaded during the examination.Tis section of the article presents literature on various techniques used for the classifcation of network trafc and analysis of the same.Tis presents the possibility of building such a system for monitoring online examinations and detection of cheating.
Lizhen and Qusay [64] have developed a phishing detection system using machine learning classifcation algorithms, namely, random forest, logistic regression, and RNN-BLSTM.Based on what the user browses, the algorithm detects if the site trying to be accessed is a phishing site and alerts the network administrator.In a similar manner, in the case of online examinations, websites posting content that could help a student cheat could be termed "phishing" sites and could be detected if the student tries to browse through them.
A deep learning algorithm has been proposed for the detection of phishing websites in Kaddoura et al. [65] along with classical classifcation algorithms such as SVM and C4.5.Tis leads to a comparatively higher accuracy of detection in comparison to conventional algorithms.Furthermore, Chiew et al. [66], Sahingoz et al. [67], and Kaddoura et al. [68] have also explored the detection of phishing sites using machine learning approaches.Table 6 tabulates the state-of-the-art mechanisms followed for the detection of phishing using machine learning.It also provides an insight into the possibility of using these techniques to detect malpractice and malicious activity during online examinations.
Apart from trying to access content from the internet, another major form of cheating could be when another person masquerades as a student to take an online examination from another location.Tis may go undetected, especially if the examination system is automated (e.g., CourseEra) and does not involve a proctor.Section 7 of this article presents recent developments in research for the detection of IP spoofng that could be a pathbreaker in the detection of masquerades during online examinations.

IP Spoofing Detection Mechanisms
IP spoofng refers to the mechanism of using false IP addresses to impersonate another computer system.Tis is a great threat to the security and authenticity of any computer network.However, in the scenario of online examinations, IP spoofng could be even more dangerous as the actual registered student who is supposed to take the exam is impersonated by another individual who can take the examination on his behalf.Terefore, for efcient detection of cheating during an online exam, it would be very benefcial if 12 Applied Computational Intelligence and Soft Computing       an IP spoofng detection mechanism was in place.Tis section of the article presents a review of the fnal possible method that could be used to combat cheating, namely, the detection of IP spoofng.Tao et al. [89] have presented a novel technique for spoofng detection.A symmetry degree is proposed for the identifcation of incomplete sessions of data transfer and other sessions that may exhibit unusual behaviours.Furthermore, based on the method of thresholding, dynamic trafc analysis is performed using the Chinese remainder theorem.
On the same lines as Tao et al. [89], Tankard [90], Liu et al. [91], and Barford et al. [92] have proposed advanced machine learning approaches for IP spoofng and malicious behaviour detection.
Olney and Karam [93] present a novel technique of watermarking trafc being generated from a computer system by obfuscating the same at the FPGA level.Terefore, an extra layer of security is added to the data being generated from the computer system, as every system has a unique watermark which cannot be copied through spoofng.

Discussion and Conclusion
Tis article presented a comprehensive review of how the soft computing paradigm has been and can be more effectively employed for the detection of cheating during online examinations.From the study, it has been concluded that though there are a few mechanisms in place for the detection of cheating during online examinations, there is still a great possibility to enhance these mechanisms by using state-of-the-art systems for user identifcation and motion detection.
In the case of face tracking and facial recognition systems, there are models available which can perform 3D tracking of faces (Jianwen et al.) [20] or use biomarkers (Xinyu et al.) [23] to detect sudden surges of emotions such as fear and confusion.Tese systems not only provide detection of minute details of facial expression change but also give high accuracy of detection.In the case of the analysis of head posture systems, several studies have been presented in this article.Te most promising ones are the ones utilizing ResNet CNN [30,31,35], resulting in multiple feature extraction and minute change detection in head posture which may be termed as malicious activity.Moving on to the usage of eye gaze tracking systems, much literature has been studied and presented in this area too.A very high resolution of up to a change of 2.3 °has been detected [21] in systems which can be employed to create systems for cheating detection in online examinations.An unexplored domain of research in the online examination realm which is related to network trafc emanating from the systems has been explored in our article.Several systems [64,65] have been presented which clearly classify the data being generated from computer systems.A robust model for cheating detection in online exams can be created utilizing these stateof-the-art facilities.Furthermore, the area of IP spoofng detection [93,105] has also been explored and presented as one of the major techniques to detect cheating during online exams.Te issues and challenges in online learning and exams including cheating detection problems were presented by Itani et al. [110].
Terefore, a comprehensive system comprising of the aforementioned verticals can be created to efectively monitor and report malicious activity and cheating during online examinations.Tis would also serve as a robust methodology to help proctors conduct online exams and make online learning a more authentic and rewarding process.

( 3 Synthesis 4 Analysis 5 Report
ii) Abstracts of the articles are read, and only the novel ones are selected under each methodology of online cheating detection (Figure2) Extraction and categorization of articles for review (i) Full texts are downloaded under each category of online cheating detection (ii) Categorization is performed on the basis of various soft computing techniques available for the task (Figure2) Conduction of article analysis and formulation of discussion (i) An in-depth study of performance metrics of the soft computing techniques used in each category for online cheating detection is performed Presentation of discussion and fnal review article (i) Further studies on potential techniques which could be used for cheating detection under each category are performed and presented in the discussion (ii) Te performance metrics of these techniques have also been presented in the discussion 4 Applied Computational Intelligence and Soft Computing to be very useful for detecting a malicious behaviour during an online exam.Table

Theta ( 4 Figure 5 :Figure 6 :
Figure 5: Correlation between power spectral density and continuous valence bands of alpha, beta, theta, and gamma with frontal lobe and nose on the top position [22].

[ 11 ]
Novel solution based on face recognition to address identity theft and cheating in online examination A complete system for analysis of video lecture based on eye tracking Device: VLEYE comprising an eye movement recorder, a dynamic AOI module, and a video analyser Classifer: thresholding 12 subjects in survey Te fxation time rate is calculated for various texts on the screen 2018 [21] A novel camera-free eye tracking sensor for augmented reality-based on laser scanning Device: low power eye tracking sensor Laboratory set up with single

Table 1 :
Steps, outcomes, and methods of the SLR process.

Table 2 :
Comparative analysis of face detection techniques.

Table 3 :
Comparative analysis of facial expression detection techniques.

Table 4 :
Comparative analysis of head posture analysis techniques.

Table 5 :
Comparative analysis of eye gaze analysis techniques.

Table 6 :
Comparative analysis of network data and classifcation techniques.

Table 7 :
Comparative analysis of IP spoofng detection techniques.