Hellinger Distance Decision Tree (HDDT) Classification of Gender with Imbalance Statistical Face Features

Abstract

Face recognition is one of the technologies used for assets protection. Face recognition also presents a challenging problem in the field of image and computer vision and has been used for the application such as face tracking and personal identification. It also frequently used in a security system such as a security camera in airport, banks, and offices. Practically, there are problems in improving face recognition performance, particularly for gender identification. It is very difficult to differentiate the person based on face appearance from different poses, lighting, expressions, aging and illumination. Sometimes it is also difficult to identify the shape of human faces because different people have a different structure of faces. This study used image retrieved from Student Information Management Systems (SIMS)from 10 male and 43 female students who're taking MAT530. The image was then generated 12 geometric landmarks using TI nspire software. The main goal of this research is to classify the gender through the images of faces and to resolve for imbalance data using Hellinger Distance Decision Tree (HDDT) classifier. This classifier was proposed as an alternative to decision tree technique which used Hellinger Distance as the splitting criteria. The result from the validation split shows that percentage split at 40% produced the highest value of accuracy rate at 77.2727% and has the most significant value of sensitivity and specificity.