Analisis Kinerja Metode YOLOv5 dan YOLOv8 pada Deteksi Kecurangan Ujian Berbasis Foto

Saputro, Whika Cahyo (2026) Analisis Kinerja Metode YOLOv5 dan YOLOv8 pada Deteksi Kecurangan Ujian Berbasis Foto. Masters thesis, Sepuluh Nopember Institute of Technology.

Text 6025202009-MasterThesis.pdf - Accepted Version Restricted to Repository staff only Download (19MB) | Request a copy

Abstract

Ujian merupakan instrumen utama dalam evaluasi pembelajaran untuk menilai kompetensi peserta didik dan efektivitas pengajaran. Namun, integritas pelaksanaannya sering terganggu oleh praktik kecurangan, terutama pada ujian daring yang meningkat sejak pandemi COVID-19. Berbagai studi enunjukkan bahwa angka kecurangan ujian daring melonjak signifikan, bahkan mencapai lebih dari 50% hingga 70% pada kondisi tertentu. Kondisi ini menegaskan perlunya sistem pengawasan yang cerdas dan adaptif guna mendeteksi perilaku mencurigakan secara otomatis.
Penelitian ini bertujuan untuk melakukan analisis perbandingan kinerja kinerja algoritma YOLOv5 dan YOLOv8 dalam mendeteksi indikasi kecurangan ujian berbasis foto. Perbandingannya didasarkan pada hasil metrik evaluasi kuantitatif yaitu precision, recall, dan mean Average Precision (mAP) setelah kedua model diuji dengan dataset dan jumlah epoch yang sama. Evaluasi dilakukan berdasarkan metrik berdasarkan metrik akurasi, precision, recall, mAP (mean Average Precision), dan hasil inferensi untuk menilai efektivitas model dalam mendeteksi perilaku mencurigakan selama ujian berlangsung.
Hasil penelitian ini diharapkan dapat memberikan gambaran empiris mengenai keunggulan dan keterbatasan masing-masing model YOLO, serta menjadi dasar pertimbangan dalam pemilihan model deteksi objek yang paling sesuai untuk diimplementasikan pada sistem pengawasan ujian otomatis.
========================================================================================================================================
Exams are the main instrument in learning evaluation to assess student competence and teaching effectiveness. However, the integrity of their implementation is often disrupted by cheating, especially in online exams, which have increased since the COVID-19 pandemic. Various studies show that the rate of online exam cheating has risen significantly, even reaching more than 50% to 70% in certain conditions. This condition emphasizes the need for an intelligent and adaptive monitoring system to automatically detect suspicious behavior. This study aims to conduct a comparative analysis of the performance of the YOLOv5 and YOLOv8 algorithms in detecting indications of photo-based exam cheating. The comparison is based on the results of quantitative evaluation metrics, namely precision, recall, and mean Average Precision (mAP) after both models are tested with the same dataset and number of epochs. The evaluation is based on metrics of accuracy, precision, recall, mAP (mean Average Precision), and inference results to assess the effectiveness of the model in detecting suspicious behavior during the exam. The results of this study are expected to provide an empirical overview of the advantages and limitations of each YOLO model, as well as serve as a basis for consideration in selecting the most suitable object detection model to be implemented in an automated exam monitoring system.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	Deteksi kecurangan ujian, YOLOv5, YOLOv8, deteksi objek, pengawasan ujian, Exam cheating detection, YOLOv5, YOLOv8, object detection, exam proctoring
Subjects:	T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK5105.546 Computer algorithms T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK7888.3 Digital computers
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55101-(S2) Master Thesis
Depositing User:	Whika Cahyo Saputro
Date Deposited:	02 Feb 2026 04:42
Last Modified:	02 Feb 2026 04:42
URI:	http://repository.its.ac.id/id/eprint/131712

Actions (login required)

View Item