Enhancing cybersecurity in higher education institutions using optimal deep learning-based biometric verification
Mahmoud Ashraf Ragab, Bandar M. Alghamdi, Rayed A. Alakhtar, Huda Alsobhi, Louai A. Maghrabi, Ghadah Alghamdi, Sameer Abdullah Nooh, Abdullah Saad Al-Malaise ALGhamdi · Alexandria Engineering Journal · 2025
Cybersecurity is an increasingly significant issue in higher education institutions, and biometric technology offers an effective solution to enhance security measures. Biometrics mentions utilizing biological features, like facial detection or fingerprint, to prove the identity of individuals. In higher education institutions, biometrics are used to improve access control and authentication models. For instance, biometric verification was utilized for secure access to computer labs, buildings, and other sensitive areas on campus. It supports preventing unauthorized access and decreasing the risk of being or other security breaches. It could be an effective tool to improve cybersecurity in higher education institutions. However, it can be vital to implement robust security protocols and privacy protection to ensure that biometric data can be utilized securely and responsibly. So, this research paper proposes the hunter-prey optimizer with deep learning-enabled biometric verification for cybersecurity (HPODL-BVCS) techniques in higher education institutions. The HPODL-BVCS technique utilizes the DL model to accomplish biometric verification in higher education institutions. To complete this, the HPODL-BVCS technique employs bilateral filtering (BF) based noise elimination to preprocess the biometric imageries. Besides, the HPODL-BVCS technique employs the ShuffleNetv2.3 model for feature extraction purposes. Additionally, the HPO model is used for the hyperparameter tuning process. The HPODL-BVCS technique utilizes a convolutional autoencoder (CAE) model with root mean square propagation optimizer (RMSProp) for classification. The simulation result of the HPODL-BVCS approach is performed on the biometric dataset. The experimental validation of the HPODL-BVCS approach portrayed a superior accuracy value of 99.81 % over existing techniques in terms of diverse performance metrics.