MANIT: a multilayer ANN integrated framework using biometrics and historical features for online examination...

Area of Science:

• Computer Science
• Artificial Intelligence
• Educational Technology

Background:

• Online education is widespread, raising concerns about academic integrity in remote examinations.
• Traditional proctoring methods are not feasible for online settings, necessitating automated solutions.
• Ensuring the authenticity of student work in online assessments is a significant challenge.

Purpose of the Study:

• To propose and evaluate an innovative automated proctoring framework for online examinations.
• To enhance the reliability and accuracy of detecting academic dishonesty in online assessments.
• To leverage Artificial Neural Networks (ANNs) and biometrics for robust proctoring.

Main Methods:

• Development of a Multilayer ANN Integrated (MANIT) framework.
• Integration of biometric features (facial orientation, eye movement using 468 landmarks) and historical data.
• Utilizing a 3-layer Deep Artificial Neural Network with angular variation-based thresholds and a regression model.

Main Results:

• The ANN achieved Mean Absolute Error (MAE) of 1.58, Mean Squared Error (MSE) of 3.96, and Root Mean Squared Error (RMSE) of 1.99.
• The MANIT framework demonstrated 88.6% accuracy in detecting five levels of dishonesty.
• Achieved high performance metrics: 90.2% precision, 90.8% recall, and 90.4% F1 score.

Conclusions:

• The MANIT framework offers a reliable and innovative solution for automatic online examination proctoring.
• The system effectively ensures academic integrity in online learning environments.
• The combination of biometrics and deep learning provides a robust approach to detecting academic dishonesty.