Student Identification in Educational Institutions Through Automated Facial Recognition Systems

The present work is focused on the comparative study of the effectiveness of the two paradigms of machine learning, namely, the technique of deep learning and the technique of probabilistic modeling, for the automated identification of students in an educational environment through the application of facial recognition technology. This work is an attempt to address the increasing need for the identification of students in an educational environment through the application of machine learning techniques, specifically the effectiveness of two paradigms of machine learning, namely, the technique of deep learning and the technique of probabilistic modeling, for the automated identification of students in an educational environment through the application of facial recognition technology.,The study employed an experimental research design that used a quantitative study method, specifically with a dataset of 620 facial images from 31 students. The study compared the effectiveness of ResNet-50's layer-wise feature extraction technique with the probabilistic feature clustering technique used in the Gaussian Mixture Model in terms of the level of classification accuracy, the balance between precision and recall, the level of confidence, and the level of efficiency in the execution of the models.,The results showed that ResNet-50 is more effective than the GMM in the classification of the facial images, with an accuracy level of 85.48% (95% CI, 83.52-87.44%) compared to the GMM's 71.77% (95% CI, 69.12-74.42%). ResNet-50 is also more robust in dealing with the poses of the students, with a lesser drop in accuracy at 6.2% compared to the GMM's 18.7%. ResNet-50 also has a more balanced level of precision and recall, with a weighted F1-score of 0.85 compared to the GMM's 0.72, even though the GMM has a computational advantage over ResNet-50 due to its significantly lower number of parameters at 7,161 compared to ResNet-50's 25 million.,From the findings of the study, it is evident that deep learning models, like ResNet-50, perform better than other models when it comes to identifying students. In particular, with regard to educational institutions that use face recognition technology, it is evident that although Gaussian Mixture Models (GMM) may be adequate, the superior performance of ResNet-50 is justified by its increased computational complexity. The study contributes to the body of literature regarding biometric systems by demonstrating its application in an educational environment and by reinforcing that models should be considered with regard to specific conditions of the issuing institution.