NIGERIAN ETHNICITY CLASSIFICATION THROUGH FUSED FEATURES FROM MOBILENET-V2 AND LOCAL BINARY PATTERN GUIDED BY ATTENTION MECHANISM

Our face plays a vital role in many human-to-human encounters and is closely linked to our identity. Significant promise exists for the automatic recognition of facial features, opening the door to hands-free alternatives and innovative uses in computer-human digital interactions. Deep learning techniques have led to a notable increase in interest in the field of face picture analysis in recent years, especially in applications like biometrics, security, and surveillance. Due to feature overlaps and dataset under-representation, ethnicity classification in computer vision is still a difficult task, particularly for African populations. This study explores Nigerian ethnicity classification, focusing on the three major groups—Hausa, Igbo, and Yoruba—using a hybrid model that integrates MobileNetV2, Local Binary Patterns (LBP), and an Attention Mechanism. The hybrid model achieved an overall classification accuracy of 87%, significantly outperforming benchmarks, particularly in Igbo and Yoruba classifications. While the Yoruba group demonstrated the highest accuracy, overlaps between Hausa and Igbo highlight areas for refinement. This research advances the field by addressing dataset imbalances, incorporating innovative feature fusion, and improving the inclusivity of computer vision models. It has practical implications for identity verification, security, and demographic research while emphasizing the importance of culturally sensitive AI systems tailored to underrepresented populations. Future work includes expanding datasets, enhancing model architectures, and exploring interdisciplinary approaches to further refine ethnicity classification.