<![CDATA[This study aims to improve identity validation in a Flutter-based attendance system that remains vulnerable to attendance manipulation, such as buddy punching and fake GPS usage. The primary issue in the previous system was that location-based validation mechanisms were unable to ensure that attendance activities were genuinely performed by authorized users. As a system implementation and software engineering study, this research applies a multi-feature similarity approach based on face embeddings, where appearance similarity serves as the primary component calculated using cosine similarity. Supporting features include geometry similarity, quality score, color similarity, and texture similarity. The system was developed using the FAST methodology, with implementation based on Flutter, Google ML Kit for face and landmark detection, and MobileFaceNet for face embedding extraction. Testing was conducted through direct implementation trials, API testing, and black-box testing involving 655 employees using a similarity threshold of 0.80. The results from 14 testing scenarios showed that all system outputs matched the expected outcomes, resulting in 100% scenario accuracy. Compared to the previous GPS-based attendance system, indications of attendance manipulation decreased from 70 cases (10.7%) to only 1 case (0.15%). In addition, the False Acceptance Rate decreased from 12.8% to 0.2%, with an average verification time of 1200 ms. These findings demonstrate that the multi-feature similarity approach based on face embeddings is capable of improving the validity and integrity of real-time attendance data on mobile devices.]]>
Copyrights © 2026
