Keeping the Integrity of Online Examination: Deep Learning to Rescue Towhidul Islam [ 0000−0001−6933−2108] , Bushra Rafia Chowdhury [ 0000−0003−0013−6971] , Ravina Akter Youki [ 0000−0001−9504−8610] , and Bilkis Jamal Ferdosi [ 0000−0001−7391−803X] University of Asia Pacific,74/A Green Road, Dhaka 1205 {17101135, 17101140, 17101154, bjferdosi}@uap-bd.edu Abstract. Online education is growing remarkably because of its ca- pability to transfer knowledge and skills remotely. It is playing a key role in this pandemic. However, it poses new challenges that need to be addressed. The examination is an integral part of any education system to judge the learner’s depth of knowledge. In the online examination, keeping the integrity of the examination is very difficult since students and examination proctors remain in remote places. Manually proctor- ing several students continuously and consistently using webcams is a tedious process. Several unethical activities may remain unnoticed. It will be beneficial if machine intelligence can help human proctors. Ex- isting researches are either based on student identification using face recognition without considering different inconsistent activities or fully automated bypassing the intelligence of human proctors. Thus, we pro- pose a semiautomated proctoring system using machine intelligence that helps a human proctor by reporting inconsistent activities and annotated videos. Inconsistent activities during examination are done to deceive in the examination, such as copying from other resources, communicating with other people, or not being present in front of the camera. Such ac- tivities can be identified by excessive eye or lip movements or the absence of the student. The proposed system uses videos captured by a webcam and continuously verifies and logs these three activities. We experimented with two different models using Convolutional Long Short Term Memory (convLSTM) and Residual Network (ResNet50). We obtained validation accuracy of 92% and 97% respectively using convLSTM and ResNet50. Keywords: Online Examination Proctoring · Examination Integrity · Machine Intelligence · Deep Learning 1 Introduction Online learning is quite popular in this age of technology because it offers more convenience and flexibility in aquire knowledge remotely. Due to the adverse effect of COVID-19, all of the educational institutions had to shut down. The only