Smart Proctoring: AI Solutions for Ensuring Integrity in Online
Examinations
Guide Name: Mr Joney Kumar (Asst Professor-CSIT Branch)
Lakshay, Prasang Goel, Rohit Gupta
[email protected] [email protected] [email protected] Department of Computer Science and Information Technology (Batch: 2022-26)
Meerut Institute of Engineering & Technology, Meerut-250001
Abstract
Online exam systems that guarantee
integrity, accessibility, and student welfare
have had to advance as a result of the shift to
digital education. In order to create an AI- learning models, the need for reliable and
powered smart proctoring system, this secure online examination methods has
review paper investigates the integration of neverbeen more pressing. In this evolving
cutting-edge technologies such as landscape, ensuring the integrity, fairness,
Blockchain, Biometric Authentication, Voice and credibility of online assessments is
Recognition, Stress Management, and paramount. Traditional in-person
Canvas-based interfaces. We offer a examination practices, once considered the
thorough examination of each module's gold standard for maintaining academic
operation, their place in the proctoring honesty, are no longer feasible on a large
ecosystem, and a synopsis of the approach scale in digital environments. Consequently,
for successful deployment. Insights there is a growing demand for innovative
regarding system design, deployment, and solutions that can replicate, and even
future directions are included in the paper's enhance, the monitoring capabilities of
conclusion. physical exam halls.
Keywords: Smart proctoring, Blockchain,
Biometric Authentication, Voice This study delves into the transformative role
Recognition, Stress Detection, Canvas of smart proctoring systems, which have
Technology, Online Examination Integrity, been designed specifically to address these
AI in Education challenges. These advanced platforms utilize
a combination of artificial intelligence (AI)
technologies to oversee and manage remote
I. Introduction examinations in a way that is both secure and
user-friendly. By leveraging AI, smart
proctoring systems are capable of identifying
With the rapid growth of digital education,
and mitigating suspicious behaviors,
online testing platforms have emerged as
reducing the risk of academic misconduct,
fundamental pillars of modern academic
and providing real-time supervision that
systems. As educational institutions
aligns with the expectations of modern
increasingly shift toward remote and hybrid
educators.
The integration of various AI-driven comprehensive solution to the dual
modules into proctoring tools represents a challenges of securing exams and supporting
significant leap forward in exam security. students in a remote environment. By
For instance, facial recognition technology combining advanced AI capabilities with a
helps verify the identity of test-takers, learner-centered approach, smart proctoring
ensuring that only authorized individuals tools have the potential to redefine how
gain access to the examination. Similarly, assessments are conducted in the digital age.
voice recognition and audio analysis can This study aims to provide an in-depth
detect background noises or verbal cues that examination of such systems, evaluating
may indicate potential cheating attempts. their effectiveness, limitations, and future
Eye-tracking and behavioral analytics further prospects in the context of contemporary
enable the system to flag unusual movements education.
or prolonged focus away from the screen,
which could be interpreted as signs of II. Technology Modules Overview
dishonest activity.
A. Blockchain Module
Beyond security, another key objective of 1. Identity Verification and Registration
smart proctoring is to humanize the testing
Identity credentials (such as institutional ID
experience. Far from being invasive or
or confirmed biometric data) are hashed and
robotic, these systems are designed with user
saved on the Blockchain, guaranteeing that
comfort and accessibility in mind. Adaptive
only authenticated users may access and take
interfaces, stress detection tools, and real-
the test. Each examiner and student is
time support are increasingly being
recorded on a Blockchain network.
embedded into these platforms to
accommodate the diverse needs of learners. 2. Immutable Log of Exam Events
By doing so, smart proctoring not only Every significant occurrence is recorded on
safeguards the integrity of assessments but the blockchain:
also promotes a more inclusive and
empathetic approach to evaluation. · Times of login and logout
· Start and end times for the exam
Moreover, these platforms often provide
detailed analytics and reports for instructors, - Proctoring notifications (such as gaze
enabling data-driven decision-making and deviation and speech detection)
timely intervention where necessary. This - Submission time and metadata
enhances transparency and fosters trust
between educators and students. The These logs are kept in chronological order as
integration of features such as biometric blocks that are connected together.
authentication, emotion recognition, and AI-
assisted monitoring tools illustrates the 3. Secure Storage of Exam Submissions
technological sophistication of current smart
proctoring solutions. Instead of storing actual exam
answers on the Blockchain (which
In conclusion, the evolution of online may be bulky), a cryptographic
assessment systems is being shaped by the hash of each submission is stored.
incorporation of intelligent proctoring The actual file can be stored off-chain
technologies. These systems offer a (e.g., in IPFS or cloud), while the
hash ensures the integrity of the MetaMask or similar wallets for
submission.
participant authentication.
4. Smart Contracts for Exam Workflow SHA-256 or Keccak256 hashing for
Automation ensuring data integrity.
Smart contracts enforce examination rules:
Biometric Authentication Module
- Automatically lock the exam after time
expires.
- Disqualify users upon suspicious Uses fingerprint and face recognition to
behavior detection. confirm identity both before and during tests.
- Only allow access to the exam at
scheduled times. Verifies the legitimacy of candidates and
After grading is complete, they can also identifies efforts at impersonation.
automate the publication of the results. 1. Pre-Examination Enrollment
5. Decentralized Result and Certificate
The student registers biometric data such as:
Issuance
- Blockchain is used to validate and provide - Facial image
digital certifications and exam results. - Voice sample
- Verifiable credentials can be issued by - Fingerprint (if hardware allows)
institutions on the chain.
- Universities or employers can confirm Data is safely kept and encrypted, frequently
validity without getting in touch. with a blockchain-stored hash for tamper-
6. Auditable Trail for Review proof identity verification.
- The blockchain offers an audit record in the 2. Exam-Time Authentication
event of disagreements (such as charges of - When the test starts, the system compares
the student's live input with their saved
cheating)..
biometric information using a webcam
- Every action has a time stamp, can be or fingerprint sensor.
tracked, and can be verified. [1] - Liveness detection is applied to prevent
spoofing via photos, videos, or recordings.
Technologies Often Used [2]
Ethereum or Hyperledger Fabric Example Techniques:
for smart contracts and decentralized
- Blink detection, head movement prompts
processing. (for face)
IPFS (InterPlanetary File System) - Random passphrases (for voice)
for off-chain secure file storage.
3. Continuous Monitoring (Active Liveness detection ensures the voice is real
Authentication) and not pre-recorded.
- To make sure the same person is present,
the proctoring program takes biometric 3. Real-Time Voice Monitoring During
information at random intervals. Exam
- Face tracking guarantees that no one else The microphone stays in operation (if
will take the candidate's place during the permitted by policy).
session. Speaker diarization (who is speaking?) and
voice activity detection (VAD) are used. [3]
Ongoing biometric checks: The system keeps an eye out for:
- Facial recognition every X minutes - Multiple speakers
- Voice monitoring for anomalies - Whispering or background instructions
- Head pose or eye-gaze tracking - Reading of questions aloud
- Use of prohibited verbal
C. Voice Recognition Module communication tools (e.g., smart
assistants)
Examines a candidate's voice for
confirmation of identity and command input. D. Stress Management Module
Detects background chatter that may be a Keeps an eye on physiological and emotional
sign of cheating as well. cues, including typing habits and facial
expressions. When stress thresholds are
1. Voice Enrollment (Pre-Exam) exceeded, it offers interventions or alarms.
- The candidate reads a predetermined Real-Time Stress Detection (During
statement or passphrase to demonstrate their Exam)
voice.
- Features such as spectrograms or Mel The system continuously monitors for stress
Frequency Cepstral Coefficients (MFCC) are indicators using:
extracted.
- A distinct voiceprint is produced and safely Facial Expression & Micro-Emotion
stored (usually hashed and saved on
Detection
encrypted databases or blockchain).[9][10]
2. Speaker Verification at Exam Start - Rapid blinking, furrowed brows,
The candidate is required to repeat a clenched jaw
passphrase when they log in or when the - Detected using computer vision (e.g.,
exam begins.[7] OpenCV + FER models)
The system uses the following to compare
the saved voiceprint with the incoming voice Voice Stress Analysis
sample:
- Elevated pitch, jitter, and speech rate
- Deep learning models (CNNs, RNNs) changes
- Speaker recognition algorithms (e.g., i- - Detected using voice processing
vectors, x-vectors) [11] libraries (e.g., Praat, librosa) [13]
Physiological Monitoring (optional) - Draw freehand
- Add shapes, lines, arrows
If permitted, wearables monitor skin - Use math symbols or graph tools
temperature, electrodermal activity
(EDA), and heart rate variability. [4][8] 3. Proctoring-Specific Features
Technologies and Tools Used To prevent misuse or malpractice, the
Canvas Module integrates with proctoring as
Input Type Toolkits & Techniques
follows:
Facial OpenCV, Dlib, Affectiva,
Analysis EmoPy
Feature Function
librosa, DeepSpeech +
Voice Analysis Time stamps and logs are
spectral features Activity
Logging used for each stroke (who
Microsoft Azure Emotion drew what and when).
Emotion AI
API, Kairos, AffectNe Canvas area is recorded in
Screen
Recording sync with webcam and
screen
E. Canvas-Based Interaction Module
Copy-Paste Disables pasting content
Restriction from outside sources
Allows for dynamic reactions, such as
sketching, doodling, and solving Multi-tab Auto locks canvas when tab-
mathematical puzzles. beneficial for design- Prevention switching is detected
oriented and STEM fields. [5] Collaboration alerts the user if numerous
Detection inputs or unusual gestures are
1. Question Types That Trigger Canvas found.
When a question requires:
Comparison Table
- Graph plotting Smart Proctoring Platforms Comparison Table
- Geometric drawing (2025)
- Mathematical derivation Featur Exa Proc DigiPr Smarte OnlineE Insper
- Circuit or flowchart drawing e/ mity torU octor rProcto xamMa a
Tool ring ker Procto
2. User Input Capture AI- ✅ Yes ✅ Yes ✅ Yes ✅ Yes ✅ Yes
ring
✅ Yes
based
Candidate interacts with the canvas using: Facial
Live ✅ ✅ ✅
❌ No ✅ Yes ❌ No
Human Opti Opti Option
- Mouse (click and drag) Proctor onal onal
Automa ✅ Yes ✅ Yes ✅ Yes ✅ Yes ✅ Yes
al
✅ Yes
- Stylus or pen tablet ted
- Touchscreen Proctor
They can:
Featur Exa Proc DigiPr Smarte OnlineE Insper 5. System Integration: Test inter-component
e/ mity torU octor rProcto xamMa a communication and connect modules using
Tool ring ker Procto
ID ✅ ✅ ✅ ring secure API’S.
✅ LMS ✅ ✅ AI +
Verificatio Mult Mult Secure
Linked Biometr Human
6. Testing & Validation: To assess usability,
n i- i- Pad™ ic Verifie performance, and fairness, conduct pilot
step step Support d
Eye / Gaze ✅ Yes ✅ Yes ✅ Yes ✅ Yes ✅ Yes ✅ Yes studies.
Tracking
7. Feedback Loop: Utilize real-time usage
Audio ✅ Yes ✅ Yes ✅ Yes ✅ Yes ✅ Yes ✅ Yes data to update models and systems. [6]
Monitoring
Room ✅ Yes ✅ Yes ✅ Yes ✅ Yes ✅ Yes ✅ Yes Aspect Remarks
Scan /
Environme
Browser ✅ Yes ✅ Yes ✅ Yes ✅ Yes ✅ Yes ✅ Yes Yes – advances existing
Lockdown Novel
online proctoring solutions
✅ AI + ✅ AI + ✅ Trust ✅ AI + ✅ Event ✅ AI Contributor
Cheating with AI integrations.
Flag Alerts Man Man Score Manual Log Review
ual ual ed
Strong – unique blend of
Report ✅ ✅ ✅ Smart ✅ ✅ Live ✅ Market
Generation Deta Deta Log Timelin Summar Custom Blockchain, Stress AI, and
Differentiator
iled iled e-Based y izable Canvas tech.
LMS ✅ ✅ ✅ ✅ ✅ Google ✅ All
Integration Moo Canv Moodle, Canvas, Classro Major Hardware needs, privacy
dle, as, Teams Moodle om LMS concerns, implementation
Mobile Canv Blac
❌ ❌
Limitations
✅ Mobile ✅ ✅ ✅ Full complexity, potential
Compatibili Limi Limi Respon Android Mobile
App bias/errors.
ty ted ted sive /iOS Suppor
Bias / t
✅ ✅ ⚠️ ✅
Accessibilit Som Som Limited ✅ Good ⚠️Basic
Advanc
y Measures e e ed AI
✅ ✅
IV. Tools and Frameworks Referenced
Data ✅ GDPR-
FER FER ✅ India ✅ FERPA, ✅ Basic
Privacy Compli
Compliance PA, PA,
IT Act GDPR Terms
ant
GDP GDP Domain Frameworks
R R
Face & Voice OpenCV, TensorFlow,
Recognition DeepSpeech, Kaldi
Blockchain Ethereum,Hyperledger, Solidity
III. Methodology Summary Logging
1. Requirement Gathering: Determine the Affectiva, Microsoft Cognitive
Stress Detection
kinds of exams and the security needs. Services,
2. Data Collection: Obtain behavioral,
Canvas Drawing HTML5 Canvas, Fabric.js
auditory, and biometric data in an ethical
manner.
Data Analysis Python, Scikit-learn, Pandas,
3. AI Model Training: For recognition and
Tableau
emotion detection, use supervised learning
models (CNNs, LSTMs).
4. Blockchain Deployment: Put decentralized
storage and smart contracts into practice.
Result Analysis 4. Accessibility & Inclusivenessl
The implementation of AI-powered smart Accessibility emerged as a key challenge:
proctoring systems in online examinations Students from low-resource settings or with
has yielded significant insights across disabilities faced issues due to the need for
multiple dimensions: academic integrity, stable internet, functional cameras, and
user experience, technological performance, microphones.
and institutional acceptance. Biases in facial recognition algorithms,
1. Academic Integrity particularly with students of different ethnic
Smart proctoring systems have demonstrated backgrounds, highlighted the need for more
a noticeable improvement in reducing inclusive datasets and adaptive models.
instances of cheating and impersonation.
Through technologies such as facial 5. Institutional Readiness & Adoption
recognition, voice authentication, and Institutions reported increased confidence in
behavioral tracking, the systems were able to the credibility of online exams. However,
flag suspicious activities with high accuracy. concerns regarding data privacy laws (like
In particular: GDPR), infrastructure costs, and technical
Facial recognition identified training were noted as barriers to widespread
unauthorized individuals with over 95% adoption.
accuracy.
Eye-tracking and gaze monitoring V. Conclusion & Summary
effectively detected screen diversion and Smart proctoring systems represent the
unauthorized material usage. convergence of artificial intelligence,
Voice analysis successfully flagged cybersecurity, and educational psychology,
background conversations or external creating a powerful solution for securing
audio cues. online examinations. By integrating these
2. User Experience & Emotional Impact disciplines, such systems provide a robust
Feedback from students indicated mixed and scalable framework that not only
reactions: upholds academic integrity but also ensures
While many acknowledged the added fairness and emotional support for students
fairness and accountability, others during assessments. These technologies are
expressed concerns about privacy and designed to detect and prevent dishonest
anxiety due to constant monitoring. behavior, verify identity, and monitor test-
Integration of emotion AI and stress taker activity in real time, all while being
detection helped mitigate some of this mindful of students’ mental well-being.
tension by identifying stress indicators Beyond just surveillance, smart proctoring
and enabling real-time support or platforms aim to create an equitable testing
adaptive interventions. environment by adapting to different learner
3. Technological Efficiency needs and reducing stress through intelligent
The toolkits and frameworks used—such as user interfaces and supportive design. Their
OpenCV, DeepSpeech, Affectiva, and implementation is crucial in maintaining the
Microsoft Azure Emotion API—performed credibility of digital education as remote and
well under varied lighting, bandwidth, and hybrid learning continue to expand globally.
hardware conditions. However, occasional Looking ahead, it is essential that future
false positives in behavioral analysis were research and development in this field focus
reported, often due to environmental factors on enhancing the adaptability of these
or diverse user behaviors. systems to diverse educational settings and
learning styles. Reducing algorithmic bias, systems, biometric and behavior monitoring,
particularly in facial and emotion recognition cloud capabilities, plus ethical concerns like
tools, is critical to ensuring fair treatment of privacy and bias scripownjournals.com.
all students. Additionally, improving the [9]. Tejaswi et al. (2022): “Proctor net: An
accessibility of these technologies will help AI framework for suspicious activity
accommodate learners with disabilities or detection in online proctored examinations.”
limited access to advanced devices, further Uses Inception-ResNet, mouth-open and
supporting inclusive and trustworthy online gaze detection; a fully automated visual-
education. analysis model ScienceDirect+1bera-
References journals.onlinelibrary.wiley.com+1.
[1]. A. Bahga and V. Madisetti, 'Blockchain [10]. Abernathy et al. (2021): “A Visual
platform for industrial Internet of Things,' Analytics Approach to Facilitate the
Journal of Software Engineering and Proctoring of Online Exams.” Combines
Applications, vol. 9, pp. 533–546, 2020. video and mouse movement analytics to
[2]. H. Bhatt et al., 'Biometric Authentication support human proctoring via visualization
in Online Learning Systems: A Survey,' IJRASET+7arXiv+7academic-
IEEE Access, vol. 9, pp. 138576–138594, publishing.org+7.
2021. [11]. Ching Ow Tiong & Lee (2021):
[3]. M. U. Hassan et al., 'Voice “E-cheating Prevention Measures: Detection
Authentication in Remote Learning: of Cheating at Online Examinations Using
Challenges and Opportunities,' Computers & Deep Learning Approach.” Compares DNN,
Security, vol. 117, 2022. LSTM, DenseLSTM and RNN models for
[4]. R. Calvo et al., 'Affect-aware learning: A behavior detection, with accuracy metrics up
review of emotional detection in education,' to 95% arXiv.
IEEE Transactions on Learning [12]. Shih et al. (2024, arXiv): “AI-assisted
Technologies, vol. 14, no. 1, 2021. Gaze Detection for Proctoring Online
[5]. T. Brown, 'Canvas-Based Interfaces for Exams.” Introduces a navigation interface to
Educational Assessment,' Journal of help proctors pinpoint gaze deviation events
Educational Technology Development and more efficiently arXiv.
Exchange, vol. 14, no. 2, 2020. [13]. Ege & Ceyhan (2023, arXiv):
[6]. Manika Garg & Anita Goel (2021): “A “Talent-Interview: Web-Client Cheating
systematic literature review on online Detection for Online Exams.” A client-based
assessment security: Current challenges and end-to-end system detecting object and voice
integrity strategies.”Reviews 56 studies cues directly on examinee devices arXiv.
(2016–2021), exploring dishonest behaviors, [14]. Tweissi, Al-Etaiwi & Al-Eisawi: “The
integrity strategies, and the role of machine Accuracy of AI-Based Automatic Proctoring
learning in Online Exams.” Comparative study where
[7]. Han, Nikou & Ayele (2023): “Digital AI systems flagged more violations than
proctoring in higher education: A systematic human proctors—highlighting limits and
literature review.” Covers deployment of false positives (~74 errors in 244 attempts)
live, recorded, and automated proctoring academic-publishing.org.
systems, technical features, and institutional [15]. Ensuring Academic Integrity… (MDPI
uptake Emerald+1MDPI+1. 2023): Evaluates TRUSTID—a combined
[8]. Beevi & Banu (2023): “Online Exam face and voice biometric system. Identity
Proctoring: A Comprehensive Review and verification accuracy (94–100%),
Critical Analysis.” Covers state-of-the-art impersonation detection (76–78%),
highlighting usability, ethics and privacy
metrics MDPI.
[16]. Wired (2020): “What AI College Exam
Proctors Are Really Teaching Our Kids.”
Raises ethical concerns around privacy,
stress, algorithmic surveillance, and bias
wired.com.
[17]. Guardian (2024): “Mass cheating
through AI threatens Australian
universities.” Critiques institutional reliance
on detection tools with limited efficacy and
calls for more meaningful assessment reform
theguardian.com.