Abstract: Manual attendance management in educational institutions and workplaces is inherently time-consuming and inefficient. On the contrary, existing automated attendance systems based on Bluetooth Low Energy, GPS geofencing, facial recognition, and conventional Wi-Fi fingerprinting often suffer from poor indoor localization, environmental sensitivity, high computational cost, and privacy concerns. This thesis proposes a hybrid Wi-Fi RSS-based attendance management system that integrates indoor fingerprinting with crowdsensing to determine whether users are inside a designated area without relying on precise coordinates, fixed signal thresholds, or biometric data. The proposed algorithm clusters real-time crowdsensed Wi-Fi RSS vectors, computes cluster centroids, and matches them with a reference radio map using k-nearest neighbors and majority voting to assign location labels. The framework was implemented and evaluated in the Department of Computer Science and Engineering, BUET, using existing campus Wi-Fi infrastructure. Experimental results demonstrate a attendance classification accuracy of 94.74%. Robustness analysis under progressive access point failures shows that the proposed approach consistently outperforms conventional Wi-Fi fingerprinting and threshold-based methods. The framework is scalable, privacy-preserving, and readily applicable to classrooms, offices, seminar rooms, and other indoor attendance scenarios. 

Presenter: Propa Punam (Std No. 0424058004)

Venue: Graduate Seminar Room