Application of Transfer Learning Method on Convolutional Neural Network (CNN) to Identify Genuine and Fake Diplomas

Abstract

The authenticity of diplomas plays a crucial role in maintaining the integrity of the education system and ensuring that recognized academic competencies align with an individual's actual achievements. Diplomas are not merely administrative documents, but strategic instruments in job recruitment and professional qualification assessment. However, with increasing educational mobility, document misuse through diploma forgery is becoming increasingly prevalent, potentially undermining public trust in educational institutions. Currently, the verification process is still largely carried out manually through visual inspection of document elements such as layout and stamps. The reliance on the examiner's experience makes this method vulnerable to inconsistencies and human error, especially when dealing with fake diplomas with visual qualities that increasingly resemble genuine documents. Diploma forgery is a problem that impacts the credibility of educational institutions and the validity of academic data. Manual inspection is often inconsistent and time-consuming. This study develops a model for classifying genuine and fake diplomas using a Convolutional Neural Network (CNN) with a transfer learning scheme. The performance of the ResNet50, VGG16, and MobileNetV2 architectures is comparatively analyzed. Data preprocessing included resizing, normalization, and augmentation. Test results showed the ResNet50 architecture achieved optimal performance with 92.63% accuracy, 92.16% precision, 94.00% recall, and 93.07% F1-score. The system was implemented in a Streamlit-based web application to facilitate the verification process.