Comparison of Machine Learning Algorithms for Face Classification Using FaceNet Embeddings

Abstract

In recent years, face recognition has grown significantly in importance and popularity. Google created FaceNet, a deep learning system, in 2015, and it performs very well in creating extremely precise and personalised numerical representations of faces, or embeddings. In order to swiftly and effectively identify people, this study evaluates FaceNet's effectiveness in producing face embeddings and applies it to a variety of classification techniques, including support vector machine (SVM), decision tree, random forest, and k-nearest neighbours (KNN). A dataset with a wide range of positions, facial expressions, and lighting settings is used for the assessment. The findings of the experiment demonstrated that SVM with an radial basis function (RBF) kernel outperformed the other assessed classification techniques, achieving the maximum accuracy of 95%. These findings demonstrate the wide range of applications that face recognition technology may be used for, including identity management and security in different settings.