Implementation of Geofencing Method and Face Recognition Integration in the Development of a Location-Based Attendance System to Improve Attendance Accuracy

Abstract

Location-based attendance systems still face challenges in validating location and verifying user identity, making them vulnerable to fake GPS manipulation and unauthorized attendance practices. This research aims to develop a web-based employee attendance system that integrates Geofencing, Haversine Formula, Face Recognition, and Cosine Similarity methods to improve the accuracy, security, and reliability of employee attendance recording at PT. Pangan Merah Putih Bengkayang. Geofencing is used to establish a virtual attendance area boundary with a 50-meter radius from the company's coordinates, while the Haversine Formula calculates the distance between the user's GPS coordinates and the central point of attendance. Face Recognition, based on the Face-api.js model, generates a 128-dimensional face embedding that is matched using Cosine Similarity with a threshold of 0.92. The system was developed using PHP, JavaScript, and MySQL, utilizing a software engineering approach and quantitative methods. Test results show that location validation using Geofencing and the Haversine Formula achieved a 100% success rate: out of five test coordinates, the system accepted four data points within the radius as valid and rejected one data point at a distance of 55.05 meters as invalid. Face recognition testing under seven conditions (lighting variations, facial angles, and unregistered faces) yielded cosine similarity values ​​ranging from 0.92 to 0.98 for registered faces. Overall system evaluation using a Confusion Matrix on 30 scenarios of location and face combinations yielded 14 True Positives, 13 True Negatives, 2 False Positives, and 1 False Negative, with an Accuracy of 90%, Precision of 87.5%, Recall of 93.33%, and an F1-Score of 90.32%. Black box testing of eight key features (login, face registration, location validation, face verification, attendance history, and logout) demonstrated that all features performed as expected. The integration of these four methods ensures that attendance can only be performed in designated areas and by users whose identities are verified, resulting in more accurate and accountable attendance data than conventional attendance systems. However, location validation still depends on the accuracy of the user's device's GPS signal and is not yet equipped with a mock location/GPS spoofing detection mechanism, while Face Recognition accuracy is still influenced by variations in lighting and the angle at which the facial image is taken