Aplikasi Absensi Pengenalan Wajah dengan Menggunakan Algoritma YOLOv11

Abstract

Manual or semi-manual attendance recording may lead to recap errors, delayed reporting, and misuse such as proxy attendance. This study develops a web-based attendance prototype leveraging You Only Look Once version 11 (YOLOv11) to perform face detection and identity recognition within a single end-to-end pipeline. The research stages include a literature review, data acquisition and pre-processing (640×640 letterbox resize and normalization), transfer-learning-based model training, and system implementation using Laravel and MySQL integrated with a Python inference service exposed via a REST API. Model performance was assessed using standard detection metrics (precision, recall, mAP@0.5, and mAP@0.5:0.95), complemented by black-box functional testing of core application modules (enrollment, attendance logging, and reporting). Internal evaluation demonstrates strong performance with precision of 0.982, recall of 0.975, mAP@0.5 of 0.987, and mAP@0.5:0.95 of 0.963. Nevertheless, performance degrades under challenging real-world conditions (extreme low-light, backlight, mask usage, and partial occlusion) and on external dataset testing, suggesting sensitivity to domain shift. Overall, the proposed system indicates practical potential for real-time attendance automation and reduced recording errors, while highlighting the need for richer, more diverse training data and cross-domain evaluation to improve generalization.