An Empirical Study on Investigating Engagement and Integrity Monitoring in Online Education Systems Using Deep Learning Models
Loading...
Date
2025-12-03
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Department of Computer Science & Engineering, NITS
Abstract
Online learning has transformed education in ways that benefit both students and
teachers. The biggest advantage of online learning is its flexibility. Students can study
at their own pace. Unlike traditional classes that require everyone to be in the same
place at the same time, online learning adapts to people’s real-world schedules. Online
education is often much cheaper than traditional classroom learning. Though the rapid
shift of online education has provided various advantages to the teacher and student,
there are certain challenges related to student engagement, academic integrity, and be
havioral monitoring during online classes/ examinations. State-of-the-art models are
mostly complex, computationally intensive, and infeasible to implement in real-time.
Moreover, there are very few approaches that provide a comprehensive analysis of student behavior throughout the examination process. Traditional proctoring systems, on
the other hand, rely primarily on manual observation or rule-based mechanisms, which
tend to be intrusive, inefficient, and constrained in terms of accuracy and scalability.
To address these challenges, this thesis proposes an intelligent, comprehensive frame
work that utilizes deep learning and multi-modal data analysis to ensure integrity while
also assessing student behavior and engagement levels in real time. Firstly, a multi
modal student engagement detection system is proposed that integrates a lightweight
deep learning model with lip movement detection to recognize the facial emotion ex
pressions and active speaker detection. The deep learning model developed in this
process is designed with fewer parameters, making it computationally efficient while
maintaining high accuracy. The effectiveness of the multi-modal system has been vali
dated using benchmark datasets such as FER-2013, JAFFE, KDEF, and RAF-DB for
facial emotion and online videos for active speaker detection. Performances have also
demonstrated reliable performance, thereby supporting a more comprehensive assess
ment of student engagement.
To understand the students’ behavior during online education, a behavioral analy
sis model is proposed that combines multiple behavioral cues. It incorporates facial
expressions, head pose, object detection, and lip movement to detect patterns that
signify focus, hesitation, distraction, or potential malpractice during online exams. In
addition, Keyframe extraction is introduced to reduce the computational load and to
enhance the performance of the deep learning model to be lightweight on facial emotion
recognition. The pre-trained deep learning models are used to analyze other behavioral
activities such as head pose, lip movement, and object detection. The effectiveness of
the behavioral model has been evaluated using benchmark and online video recordings
to enable real-time performance without compromising accuracy. A teacher’s dash
board developed in this process aggregates and visualizes the analyzed data, offering
actionable insights into student behavior, engagement levels, and integrity violations
for the teacher’s reference.
Finally, to maintain online exam integrity, an intelligent proctoring model is proposed,
which provides a friendly environment for students and proctors. It combines academic
misconduct, stress detection, and enabling an adaptive monitoring proctor interface in
a single intelligent system to create a balance between integrity, fairness, and student
well-being. The proposed model will perceive academic misconduct activity and other
normal activities, such as writing the exam, being distracted by looking somewhere,
etc. Specifically, it includes designing and implementing a deep learning model capable
of real-time monitoring of student activity and facial expression. In addition, in this
work, the analysis of facial expressions is used to propose an innovative approach to
identifying student stress levels, and when they are highly stressed, the system will
give motivational feedback to improve concentration and performance. The proposed
model has been evaluated using real-time data from a student who was writing the
online examination