Facial Palsy Image Classification Using Landmarks Extraction and Extreme Learning Machine

Facial palsy is a condition characterized by facial paralysis, affecting the patient's motor function. Detection is typically done through a clinical expert's direct observation of facial muscles. However, manual examinations are less effective as patients tend to go for health checks when symptoms indicate severe dysfunction. Due to these challenges, a classification system is built as early screening for facial palsy. This research aims to propose the detection of facial palsy using Extreme Learning Machine algorithm. In this research, we use three datasets: CK+, JAFFE, and Youtube Facial Palsy (YFP). The dataset is divided into training and testing data in an 80:20 ratio. Applied preprocessing techniques of landmark facial feature extraction, selection, and Euclidean distance calculation at specific face points. We employ k-fold cross-validation to find the optimal parameter during the training process. We also implemented oversampling and undersampling techniques to handle imbalanced data. The system classifies facial palsy into normal, mild, moderate, and severe dysfunction. The ELM model achieved Accuracy values of 0.9949, Precision of 0.9888, Recall of 0.9887, and F1-score of 0.9888 on balanced data using an oversampling technique. The current research indicates that the implemented techniques can handle imbalanced data, and contribute to better model performance.