A machine learning assistant for detecting fraudulent activities in synchronous online programming exams.

Area of Science:

• Computer Science
• Artificial Intelligence
• Educational Technology

Background:

• Online learning expansion presents academic integrity challenges, especially in remote exams.
• Existing proctoring systems often rely on physical monitoring and human review, which is inefficient.
• Students must remain online during exams, creating opportunities for academic misconduct.

Purpose of the Study:

• To develop and evaluate a machine learning-based assistant for real-time fraud detection in online programming exams.
• To improve the efficiency and accuracy of identifying academic dishonesty during digital assessments.

Main Methods:

• A machine learning model combining a convolutional neural network (CNN) and a recurrent neural network (RNN) was developed.
• The system analyzes sequences of screenshot frames from student screens during exams.
• Training involved data augmentation and class-weight adjustments to optimize performance.

Main Results:

• The system achieved 95.18% accuracy and an F2-score of 94.2%, prioritizing recall for effective cheating detection.
• Data augmentation and class-weight adjustments significantly improved model performance.
• Transfer learning and alternative loss functions did not yield further improvements.

Conclusions:

• The machine learning assistant effectively aids instructors in detecting fraudulent activities in real-time during online programming exams.
• Instructors reported high satisfaction with the system's ability to rapidly identify cheating.
• The study highlights the potential of AI for real-time monitoring in large-scale online assessments.