In our prior chapters, we discussed image preprocessing, landmark localization, feature extraction, and feature selection. This chapter presents a systematic overview of 2D image-based facial attractiveness prediction methods, including machine learning and deep learning approaches. While most existing deep learning models treat facial beauty analysis as a standard classification task, they often overlook prior knowledge from traditional methods, particularly the importance of geometric features. To address this, we highlight a geometric prior-guided hybrid deep network that integrates both global and local cues. Landmarks of the face and facial organs are used to extract geometric features, while a dual-branch architecture is employed: one branch with a Swin Transformer captures full-face global patterns, and another with a residual network focuses on local regions of interest. A multi-scale fusion module further enhances complementary information between the two branches. A hybrid loss function, including geometric regularization via landmark regression, enforces the embedding of geometric constraints into learned features. Finally, we present comparative experiments on the SCUT-FBP5500 and SCUT-FBP datasets, with analyses of performance and applicability, offering insights to guide future research in facial beauty prediction.