Toward Sustainable Online Education: AI-Powered Hybrid Proctoring with LSTM and CNN-Based Anomaly Detection to Enhance Academic Integrity

This 21st century is the era of e-education, and there is an urgent need to adapt to the modern needs of education in line with the 4th Sustainable Development Goal (SDG). The goal of this study is to align with SDG 4's goal of providing inclusive and equitable quality education. It aims to discuss the emergence of proctoring software that utilises artificial intelligence as a means of addressing the increasing cases of cheating in remote learning environments and online assessments, thereby reducing the need for physical infrastructure and travel, and contributing to sustainability. This study proposes a hybrid proctoring system based on two different folds: the first fold detects the significant improvement in the candidate's marks from the custom dataset of 350 students to identify the suspected candidates using Long-Short Term Memory (LSTM), and the second fold analyses the exam video recording of suspected candidates frame by frame to perform the behaviour analysis to detect the anomalies with the help of Convolutional Neural Network (CNN). In this paper, various anomalies were identified, including off-screen gazes, the use of cell phones and earphones, and talking. The proposed system obtains an accuracy of about 87.8% as well as exhibits resource-efficient performance with respect to processing time, CPU usage, along with memory usage.