<![CDATA[Automated attendance systems based on facial biometrics can reduce manual recording, but their deployment requires a clear separation between face detection and identity recognition. This study evaluates an Eigenface-based attendance prototype in a multi-face classroom setting and reanalyzes the archived test results using a reproducible statistical protocol. The prototype processes camera frames through face localization, grayscale conversion, cropping and resizing, principal component analysis projection, Euclidean-distance matching, and attendance recording. The available evaluation comprised 30 trials, each containing 17 target faces, yielding 510 face-detection opportunities. Trial-level detection coverage was defined as the number of detected faces divided by 17. Across all trials, 310 faces were detected, corresponding to an aggregate coverage of 60.78%. Trial-level coverage ranged from 29.41% to 82.35%, with a mean of 60.78%, a standard deviation of 14.44 percentage points, and a 95% t-confidence interval of 55.39%–66.17%. These results demonstrate that the prototype can execute the intended attendance workflow but is not yet sufficiently reliable for unsupervised operational use. Because the archived data contain only detected-face counts, they cannot establish identity-recognition accuracy, false acceptance, or false rejection. The study contributes a corrected evaluation of the prototype, an explicit distinction between detection and recognition evidence, and a deployment-readiness framework for subsequent controlled validation. ABSTRAK Sistem absensi otomatis berbasis biometrik wajah berpotensi mengurangi pencatatan manual, tetapi penerapannya memerlukan pemisahan yang jelas antara deteksi wajah dan pengenalan identitas. Penelitian ini mengevaluasi prototipe absensi berbasis Eigenface pada skenario kelas multiwajah dan menganalisis ulang hasil pengujian arsip menggunakan protokol statistik yang dapat ditelusuri. Prototipe memproses bingkai kamera melalui lokalisasi wajah, konversi ke skala abu-abu, pemotongan dan perubahan ukuran citra, proyeksi principal component analysis, pencocokan menggunakan Euclidean distance, serta pencatatan kehadiran. Evaluasi yang tersedia terdiri atas 30 percobaan dengan 17 wajah target pada setiap percobaan sehingga menghasilkan 510 peluang deteksi. Cakupan deteksi setiap percobaan dihitung sebagai jumlah wajah terdeteksi dibagi 17. Secara keseluruhan, 310 wajah terdeteksi atau setara dengan cakupan agregat 60,78%. Cakupan per percobaan berkisar antara 29,41% dan 82,35%, dengan rata-rata 60,78%, simpangan baku 14,44 poin persentase, dan interval kepercayaan t 95% sebesar 55,39%–66,17%. Hasil tersebut menunjukkan bahwa prototipe dapat menjalankan alur absensi, tetapi belum cukup andal untuk digunakan tanpa pengawasan. Karena data arsip hanya memuat jumlah wajah terdeteksi, hasil ini belum dapat membuktikan akurasi identifikasi, penerimaan salah, maupun penolakan salah. Kontribusi penelitian adalah koreksi evaluasi prototipe, pemisahan bukti deteksi dan pengenalan, serta kerangka kesiapan penerapan untuk validasi lanjutan.]]>
Copyrights © 2026
