Cheating Detection in System Computer-Based Test (CBT) Based on Participant's Facial Images using PCA Feature Reduction on CNN

Covid-19 has caused whole world to implementing online learning to prevent mass gatherings in one place. By conducting online learning, exams also running with online method. Before pandemic covid-19 cheating during exams was already become problem in evaluating the result of learning and with pandemic, controls to prevent cheating are urgently needed. Cheating during exam with computer-based test (CBT) become problem, this is challenge for artificial intelligence in detecting cheating during exams based on participant's face. This paper presenting a comparison model Convolutional Neural Network (CNN) and model CNN with reduction Principal Component Analysis (PCA) in detection cheating during exams. The dataset that uses to training and testing with the CNN architecture that was built provides 77% accuracy. By adding PCA reduction to feature CNN, the accuracy increases to 83% with same training and testing data. Form this case we can conclude that using PCA in reducing CNN features can increase model performance.