Student face recognition optimization utilizing the k-nearest neighbor method and particle swarm optimization in testing models and mobile devices Iskandar, Umar Tsani Abdurrahman, Mohamad Anas Sobarnas, Nanik Wuryani
Sekolah Tinggi Teknologi Muhammadiyah Cileungsi
Abstract
In order to achieve high accuracy with a reasonable processing time for artificial intelligence-based facial recognition, the proper technique must be used. In this study, two different kinds of experiments were run on the program^s model and application sides. On the modeling side, the Linear Discriminant Analysis (LDA) approach is used for feature extraction, and the Particle Swarm Optimization (PSO) method is used for feature selection in the K-Nearest Neighbor (KNN) classification. The ORL database, which contains 357 facial photographs from 7 different people, with 51 images per person, served as the study^s dataset. The LDA and KNN algorithm models are anticipated to have accuracy increases of more than 80% in model testing. In tests, the PSO-based LDA and KNN algorithms achieved the maximum accuracy of more than 85% while processing data more quickly. Performance is increased by more than 5% as a result of modeling optimization on a PSO basis. The improvement in accuracy demonstrates that PSO is able to choose the best attributes that are suitable and practical to analyze during categorization. In addition to testing the application program, it is envisaged that modeling will be able to identify the suitable model device on a real system by measuring the response time of objects known as devices.
