Machine Learning Computational Design on the Edge Device in IoT Network to Classify Audience Face Image Attention

The limitations of edge devices in processing capacity, storage, and memory can constrain the computational complexity of edge computing in extracting information directly from data sources. This information is then forwarded by the edge gateway to servers or the cloud, or to other edge devices, via the Internet of Things (IoT) network for further data processing. A model in machine learning has a mathematical structure algorithmically, which represents complex computations in the implementation of artificial intelligence to automate problem-solving. In this research, with the Cross-Industry Standard Process for Data Mining (CRISP-DM) methods, the proposed design is a machine learning computing system on edge devices, using a Convolutional Neural Network (CNN) based model, specifically the MobileNetV2 architecture. The edge computing process with machine learning occurs on the ESP32-CAM, which is connected to a cloud storage system via the Internet of Things (IoT) network. Machine learning computation on the ESP32-CAM is particularly advantageous for the classification of audience facial images during teaching and learning activities, that Captured by the ESP32-CAM, the images are categorized with labels of focused and unfocused. The image data, including its labels, is then sent to cloud storage. The dataset used in building the model consists of 94 training data and 35 testing data. Model testing was conducted 4 times, with a batch size of 256 and a total of 80 epochs. The total number of test data is 35, with 21 out of the 35 test image data correctly classified. Thus, the results obtained are 60% for accuracy, 66% for AUC, 72% for precision, 60% for recall, and 61 % for the F1 score.