Development of an Improved Chicken Swarm-Convolutional Neural Network for Bimodal Biometric based Ethnicity Identification System

The advancement in science and technology has brought about the use of biometrics applications for establishing individual identity. Bimodal biometric systems have gained significant attention in applications where unbiased identification is required. This study presents an enhanced ethnicity identification framework by integrating Convolutional Neural Networks (CNNs) with an Improved Chicken Swarm Optimization (ICSO) algorithm using face images and fingerprints. A dataset comprising 600 subjects from three major ethnic groups in Nigeria (Yoruba, Igbo and Hausa) was obtained. Image augmentation techniques were applied to each ethnic group to improve the dataset, this increased the dataset to a total of 3,000 images, 80% was used to train while 20% was used to evaluate the CNN models optimized with CSO and ICSO using random sampling cross-validation method for each data sample group at 0.75 threshold value. The experimental results show that the ICSO-CNN model consistently outperforms both the baseline CNN and CSO-CNN models across all evaluation metrics, achieving an accuracy of 98.83%, false positive rate of 0.5%, F1 score of 98.24% and a recognition time of 90.77 seconds for Yoruba ethnicity, and an accuracy of 98.5%, false positive rate of 0.75%, F1 score of 97.73%  and a recognition time of 90.09 seconds for Igbo ethnicity while Hausa ethnicity had an accuracy of 99.17%, false positive rate of 0.25%, F1 score of 98.74% and a recognition time of 89.34 seconds. These findings highlight the effectiveness of integrating chaotic map into optimization processes to enhance CNN training efficiency and identification robustness in Ethnicity Identification.