IMPLEMENTATION OF DEEP LEARNING TECHNOLOGY IN THE ATTENDANCE SYSTEM AT PT HARPA OCEAN BERSAMA

Abstract

Employee attendance integrity constitutes a fundamental pillar of organizational management. Nevertheless, many companies encounter inefficiencies and vulnerabilities associated with manual or contact-based recording systems. PT Harpa Ocean Bersama is currently grappling with analogous challenges, as its reliance on conventional methods has resulted in administrative inefficiencies and a vulnerability to fraudulent practices, such as "buddy punching." The objective of this study is to develop a contactless Face Recognition Attendance System to automate and secure the workforce verification process. The system is constructed using a Convolutional Neural Network (CNN) framework, specifically leveraging the MobileNetV2 architecture combined with Transfer Learning. This approach was selected to optimize detection accuracy while minimizing computational costs, making it suitable for real-time application. The development methodology encompasses a rigorous pipeline, which includes the following steps: data acquisition, preprocessing using Haar Cascades for precise face isolation, data augmentation, and model fine-tuning using the TensorFlow library. The resulting system boasts robust employee identification capabilities, achieving a testing accuracy of 99.3% and a precision of 100%. This solution has been demonstrated to enhance data reliability and operational speed in comparison with conventional manual methods. It is anticipated that the implementation of this system will modernize human resource operations at PT Harpa Ocean Bersama, thereby ensuring a transparent, efficient, and fraud-resistant attendance environment.