<![CDATA[The student attendance system is useful for monitoring student attendance. The current technology is technology capable of detecting an object, such as fingerprints, voice, eye retinas, and faces. The author will create a model that can be used to detect student faces. In this study the authors used a modified Convolutional Neural Network (CNN) algorithm. The complexity of the CNN designed is in accordance with the specifications of the hardware and software used. Face data is taken directly from students in class (private dataset). Recording of students' faces using a standard quality webcam camera. The images produced by each student are 126 images with a total of 20 classes (labels). Taking pictures with various angles of the face, namely from above, below, front, left side and right side. The augmentation techniques used are flip, random rotation and affine techniques to enrich the data. Regularization techniques, such as dropout are also used. This is in order to increase accuracy, speed of model training and avoid overfitting of the built model. The evaluation results with the confusion matrix on the modified Convolutional Neural Network (CNN) algorithm produce a faster model training process with 5.31 hours and accuracy reaching 97.78%, the loss value is stable at 0.1177, loss validation with the number 0.0192, with as many iterations (epochs) as 60. The resulting model will be developed on a prototype of the student attendance system.]]>
Copyrights © 2023
