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

Online education has become a globally accepted norm, bringing benefits and challenges. One of the most debated aspects is the academic integrity of online examinations. Without physical proctoring, the authenticity of the candidates' scores can always be called into question. This paper proposed an innovative Multilayer ANN Integrated (MANIT) framework for Artificial Neural Network (ANN)-assisted automatic online proctoring using the combination of biometrics and historical features. It tracks facial orientation and eye movement using 468 landmarks. Empirically designed angular variation-based thresholds associated with a regression model developed with a 3-layer Deep Artificial Neural Network have made the proposed MANIT framework a reliable automatic online examination proctoring system. The Mean Absolute Error (MAE), Mean Squared Error (MSE), and Root Mean Squared Error (RMSE) of the ANN are 1.58, 3.96, and 1.99, respectively. The overall system detects the five levels of dishonesty in online examinations with 88.6% accuracy. The precision, recall, and F1 score are 90.2%, 90.8%, and 90.4%, respectively. The innovative design, cross-validation-based reliability, and outstanding performance of the MANIT framework have made it an excellent proctoring system to ensure academic integrity in online examinations.