ANALISIS PERBANDINGAN ALGORITMA CONVOLUTIONAL NEURAL NETWORK, FACENET DAN K-NEAREST NEIGHBORS PADA SISTEM ABSENSI BERBASIS PENGENALAN WAJAH

Abstract

This study aims to analyze and compare the performance of Convolutional Neural Network (CNN), FaceNet, and K-Nearest Neighbor (KNN) algorithms in a face recognition–based attendance system. The research was conducted using the Labeled Faces in the Wild (LFW) dataset, which was augmented to improve data variability. Each algorithm was tested using three different train test split ratios: 70:30, 80:20, and 90:10. CNN was implemented as an end-to-end classifier, while FaceNet acted as a feature extractor producing facial embeddings, and KNN served as a distance-based classifier. The evaluation applied accuracy score and mean squared error (MSE) to measure prediction accuracy and error rate. The results show that KNN achieved the best performance with the highest accuracy of 90.20% and the lowest MSE at the 90:10 data ratio. FaceNet demonstrated stable performance with accuracy ranging from 79% to 83%, while CNN showed significant improvement as the amount of training data increased. Based on these findings, KNN is considered the most optimal algorithm due to its balance between accuracy and low error rate. These results are expected to serve as a reference for developing more accurate and efficient face recognition–based attendance systems.