Performance Evaluation of Face Recognition Model in Deep Learning using Old Facial Photos: A Mathematical Modelling Using SVM Supervised Learning

Abstract

Deep Learning has been a remarkable state-of-the-art method in any classification challenge, particularly in face recognition applications. In this paper, Feature Extraction in face recognition using Deep CNNs handpicked pre-trained CNN architectures such as InceptionV3, MobileNetV2, ResNet50, and VGG19 were experimentally explored. Initially, these architectures extracted important features from eight (8) classes of face photos with large age differences of ten (10) years from the present age of an individual. The features were processed with the application of a Support Vector Machine (SVM) classifier to enhance its performance. The evaluation of each model was based on average scores ofaccuracy, precision, recall, and f1-score. The results concluded an accuracy of 84.60%, a weighted precision of 85%, a weighted recall of 84.60%, and a weighted f1-score of 84.60% obtained by ResNet50.Further, ResNet50 has the highest obtained 98% generated ROC-AUC score. With the results presented, ResNet50 isrecommended for application development related to face recognition with the consideration of large age gaps of 10 years.