Facial Paralysis Classification Based on Multi-Level Feature Fusion and Dimensionality Reduction

Conventional diagnosis of facial paralysis relies on clinicians' professional experience to visually assess facial asymmetry in patients. Such a method is prone to subjective bias, lacks precise quantification, and poses challenges for implementation in rural healthcare and telemedicine. To address these limitations, this study proposes an automated machine learning framework that integrates various facial features with dimensionality reduction techniques. It is then applied to multiple machine learning models for training and performance comparison to evaluate the accuracy of FP classification. Experimental results show that with feature fusion and dimensionality reduction, it achieved an accuracy of over 98%, especially with SVM and LR classifiers. Therefore, the proposed learning framework can be effectively applied to the early detection of facial paralysis and clinical decision support, offering long-term quantitative monitoring and demonstrating strong potential for use in telemedicine and home-based healthcare applications.