The 21st International Conference on Advances in ICT for Emerging Regions (ICTer2021) Towards a Framework for Online Exam Proctoring in Resource-Constrained Settings Focusing on Preserving Academic Integrity Dilky N. Felsinger University of Colombo School of Computing Colombo, Sri Lanka dilkyfelsinger@hotmail.com Ishani Fonseka University of Colombo School of Computing Colombo, Sri Lanka cli@ucsc.cmb.ac.lk Thilina Halloluwa University of Colombo School of Computing Colombo, Sri Lanka tch@ucsc.cmb.ac.lk Kasun Karunanayake University of Colombo School of Computing Colombo, Sri Lanka ktk@ucsc.cmb.ac.lk Abstract— In recent years, there has been a remarkable surge in the global usage of online learning. However, it is facing difficulty maintaining academic integrity during online assessments. To that purpose, many institutions have implemented remote exam proctoring through video- conferencing software, which necessitates a large number of human proctors when thousands of students take an online test at the same time. Unfortunately, government-funded institutes in Sri Lanka cannot use automated online exam proctoring solutions on the market since they are expensive and frequently do not suit low-resource contexts. This research aims to find techniques to reduce computing costs and network data consumption in automated online exam proctoring scenarios. First, we will create a dataset by manually proctoring an online exam via video conferencing because we do not have a publicly accessible dataset to capture potential acts of misbehaviour. Second, we will use the data collected to find an online proctoring system that works well under resource constraints. Finally, the study will assess the suggested method's usefulness in detecting academic dishonesty in low-resource situations. Keywords—Academic Integrity, Online Exams, Proctoring I. INTRODUCTION Global adoption of Online Learning has increased rapidly in recent years [1]. Access to quality learning material from world-renowned universities [2], the ability to learn at one's own pace [3], and even getting the ability to earn college credits [4]–[6] by engaging in Massive Open Online Courses (MOOC) are some of the benefits of online learning. Despite its advantages, Online learning faces challenges in conducting laboratory training and conducting online exams while ensuring academic integrity and protecting the credibility of the programs [1], [7]. Hayashi et al. report a considerable gap in equal access to the Internet between urban and rural areas. The affordability of the Internet is also recognized as a barrier to adopting online learning methods [7]. Though Online Learning has seen a sudden usage surge during the Coronavirus pandemic, it faces issues maintaining academic integrity during online testing [8]. Many universities have adopted remote exam proctoring through video-conferencing software [9]. However, in cases where thousands of students take an online test simultaneously, manual proctoring using video-conferencing software can be a tedious task. Therefore, this research aims to implement a framework to conduct online exam proctoring using low resources to preserve academic integrity. II. RELATED WORK There are three types of online proctoring methods that have been implemented thus far. Namely, online human proctoring, semi-automated proctoring, and fully automated proctoring. In online human proctoring, an invigilator watches the student through the software while a candidate writes the exam. Proctoring using video conferencing software belongs to this category [9], [10]. Fully automated proctoring has eliminated the human factor in online proctoring while using machine learning to identify the wrongdoers. Li et al. have presented an online proctoring method using head movement and mouse movement detection that uses deep learning [11]. Atoum et al. has used a multimedia approach using a webcam and a wearcam to detect if a test taker is cheating using a two- stage method where the first stage extracts features from the videos using a set of ensemble models and the second phase being detecting misbehaviour using extracted features using a support vector machine [12]. Semi-automated proctoring has been introduced to overcome the concerns of the fully automated proctoring methods that include the human factor and technology. In Semi-automated proctoring, the software detects cheating behaviours and patterns, and a human will examine that furthermore to make the final decision. Hence, in semi- automated proctoring, the human proctor will have to make the final decision based on his insights and findings. Li et al. have presented a semi-automated online proctoring method to detect misconduct in MOOCs that consist of three components, an Automatic Cheating Detector to invigilate students using a webcam; Peer Cheating detector to identify if the susceptible behaviour is truly academic misconduct and a Final Review Committee to decide whether the student commits plagiarism eventually [13]. Many studies have elaborated that an Automated Online Exam Proctoring system contains two main components, namely, Exam Candidate Authentication and Detection of Academic Misconduct [14], [15]. In most of the studies conducted to address the problems in online proctoring, a priority has been given to user authentication and successful methods have been