Ethnicity Classification: A Machine Learning Approach

Abstract-Recently, researchers in the field of Machine Learning have paid a lot of interest to the human face. Soft biometric features taken from a facial image can be used to distinguish between racial classes. Other soft biometrics features include race, age, gender, and emotions. Research has employed different techniques (traditional and deep learning) in predicting the major racial classes (Asian, Hispanic, African, Caucasian) with outstanding cutting-edge performances. Recently, research has focused on identifying distinguishing characteristics in sub-racial (ethnic) groups. Racial profiling has been used in a variety of fields, including social media profiling, security surveillance, law enforcement, and targeted advertising. By seeing relevant studies in the field, we noted that the Black race (African/African American) is considered a single racial entity, models developed do not have practical application in the Nigerian domain, and most of the datasets available are racially imbalanced. As a result, the goal of this research is to create a unique dataset with accurate labels for Nigeria's three major ethnic groups, and then using deep learning techniques to classify these labels. There are three labels in the image dataset: Hausa, Igbo, and Yoruba. For feature extraction and classification, a pre-trained Convolutional Neural Network (CNN) was used. The model was evaluated on the test, and The Hausa ethnic group had the highest accuracy of 87.3%; lower accuracies were recorded from the Igbo and Yoruba subclass, which gave an accuracy of 56.0% and 56.0%, respectively. The result could be attributed and migration and inter-ethnic marriages which have dwindled the boundary between the ethnic groups.