This study addresses the issue of cheating detection in cybersecurity courses by proposing an evaluation method that combines Capture the Flag (CTF) competitions with machine learning techniques. Through the analysis of a combination of features related to operational behavior, temporal patterns, and network activities, we aim to optimize feature selection and improve the accuracy and stability of models in detecting cheating behaviors. The results show that machine learning models like Random Forest and XGBoost are effective in detecting potential cheating behaviors when multi-dimensional features are combined, significantly enhancing detection performance. These findings provide a fairer and more effective solution for cheating prevention in cybersecurity education, supporting future educational development.