Automated classification of the severity of hemifacial microsomia (HFM) from craniofacial scans remains challenging due to limited medical imaging data, challenging anatomical features, and computational constraints. We propose an efficient and effective classification framework that leverages lightweight STL meshes and a streamlined processing pipeline, including mesh cleaning, edge-focused point-cloud processing strategy and intra-class MixUp augmentation. Specifically, the proposed edge-focused strategy is tailored explicitly to select points near anatomical boundaries critical for HFM diagnosis, preserving essential diagnostic features while dramatically reducing data complexity. To overcome data scarcity, we implement intra-class MixUp, significantly improving model robustness by synthesizing anatomically realistic additional samples. Evaluations on a clinically labeled HFM dataset confirm that our optimized model, built upon the Pretrained PointNet backbone, achieves superior accuracy and Macro-F \(_1\) compared to state-of-the-art alternatives. Our results demonstrate that this edge-focused, lightweight, augmentation-enhanced approach effectively balances diagnostic accuracy with computational efficiency, advancing automated craniofacial severity classification.