Detecting Face Expressions in Real-Time Using Convolutional Neural Network (CNN) Algorithm

This research discusses the use of a Convolutional Neural Network (CNN) with the MobileNetV2 model in the real-time detection of human facial expressions. This research aims to develop a human face expression detection system using deep learning algorithms. This study used the observation data collection method and obtained secondary data from the FER2013 data set which contains 28,709 training samples, 3,859 validation data sets, and 3,859 test samples, for a total of 35,887 images with a resolution of 48x48 and seven categories of facial expressions. The training results showed that the CNN model using MobileNetV2 achieved an accuracy of 57% in the training process and 51% in the validation process. Based on the analysis of these results, testing using a confusion matrix with an accuracy of 51% concluded that the model was unable to properly recognize patterns of data with disgust and fear categories, leading to low accuracy. Some factors contributing to the system's inability to recognize expressions were due to similarities between facial expressions such as sad and fearful, or sad and disgusted. This study provides new insights into the development of technology for detecting human facial expressions using deep learning and the MobileNetV2 model.