Facial Based Age Estimation Among Dark Skinned Hominid Species

By intuition humans can estimate the age of an individual effortlessly. However, this task can be very challenging for face age recognition system. These challenges primarily are attributed to the unavailability and the under-representation of appropriate dataset used in the training of age estimation models. A specific scenario is the underrepresentation of black race which raise the concern of algorithm accountability resulting in inefficient model formulation. In this study, a dataset named Faces was curated from secondary sources containing face images of the black race. A hybrid model consisting of a KNN classifier and an SVR was then proposed to estimate these ages. For a fair comparison, the trained hybrid model was experimented on three other datasets and compared to the LARR model which was the state-of-the-art model at the time of study. The results showed that the proposed model generated the least MAE on the Faces dataset which affirms our hypothesis that a model works better with the kind of data used curated.