Authors: Tanvi Mahesh Waghchaure, Professor Bangar Abhishek Popat, Professor Bhosale S.B.
Abstract: Exam paper leaks damage the fairness and credibility of examinations, and most leaks involve people who legitimately hold access to the paper, such as setters, moderators, printers, couriers and administrators. This makes the problem closely related to malicious insider threat detection and to data exfiltration control. This paper presents a focused narra-tive review of verified sources on machine learning (ML) and cybersecurity techniques that could support exam paper leak detection and prevention. It examines three connected themes: insider threat categories and detection approaches, ML and deep learning models for intrusion and anomaly detection, and preventive cybersecurity controls such as en-cryption and threshold secret sharing. The literature shows that behaviour-based, anomaly and hybrid ML methods are widely used for insider threat detection, that a hybrid explainable deep learning model reached 97.11% accuracy over a 37-class combined test set of 1,957,876 samples, and that an ensemble of five classical ML models reached accuracy, precision, recall and F1 score of 99.4%, 99.6%, 99.7% and 99.7% on a cleaned KDD Cup 1999 dataset. However, none of the reviewed sources studies exam paper leaks, the datasets are network or simulated insider data, rare classes remain hard to detect, and monitoring staff raises privacy and legal questions. The review contributes a comparative analysis of sources, datasets and reported results, a proposed conceptual seven-layer framework linking secure custody, key release control, activity logging, behaviour analysis, explainable alerts, response and audit, and a prioritised list of research gaps.
