Aquila-optimized Recurrent Neural Network for Enhanced Facial Biometric Crime-control Systems

Abstract

The increasing complexity of criminal activities and the demand for rapid, reliable identification mechanisms have strengthened the relevance of biometric-based crime-control systems. This study presents an optimized facial biometric recognition framework utilizing an Aquila-enhanced Modified Recurrent Neural Network (AORNN). A dataset of 2,160 real-world facial images was preprocessed through grayscale conversion, cropping, normalization, and histogram equalisation to enhance feature consistency. The Aquila Optimizer was employed to fine-tune the RNN parameters, improving convergence stability and classification performance. The system was implemented in MATLAB R2023a and evaluated using standard biometric metrics. Experimental results show that the AORNN achieved superior performance, attaining a recognition accuracy of 95.83% and an Equal Error Rate of 4.10%, outperforming the baseline RNN. These improvements demonstrate the model’s enhanced discriminative capability and suitability for real-time crime-control applications where accuracy, reliability, and computational efficiency are critical. Future research will focus on multimodal biometric integration, dataset expansion, and deployment on edge-computing architectures to support operational law-enforcement environments.