Application of Deep Neural Network Modifications for Face Recognition in Attendance Systems

The conventional method of collecting attendance as evidence of student attendance is considered ineffective because it consumes a lot of time and effort. The validity of the data is questionable. There have been many models that have been applied to facial recognition-based attendance systems. However, this model needs much training data so that the model's accuracy is high. In this study, a modification of the deep neural network model for the attendance system is proposed that can work on a small amount of training data. The proposed model is a modification of the DenseNet201 model with batch normalization and average pooling layer. Even though our model's training time is quite long, this model modification can achieve the highest accuracy value of about 90% compared to other pre-trained models, namely ResNet50 and MobileNet.