A Comparative Analysis of Machine Learning Algorithms for Facial Paralysis Detection.

Facial paralysis refers to the impairment or a condition where the facial muscles on one or both sides lose the ability to function, significantly affecting a person’s ability to perform essential functions such as talking, closing the eyes, swallowing, having food, and expressing emotions naturally through facial movements. The primary method for diagnosing facial paralysis is to observe the patient’s face for facial spatial information, such asymmetry. The clinical subjective assessments of therapists are still used to treat this condition. In the current era of technology, the automation for this activity is therefore highly significant and is the subject of much research in both academic and industry domains. In this paper we have used the traditional facial landmarks identification using machine learning approaches to test the accuracy of different machine learning algorithms. We used 11 selected key-points and regions from the facial image crucial for identifying facial paralysis. The detection process comprises of three stages in which the initial stage is the detection of the face, followed by the extraction of landmarks and Region of Interests (ROIs), the computing of the facial measure, and the categorization of facial paralysis. Three different data sets are used for the experiments. The experiments show that the Random Forest has the maximum accuracy of 92% as compared to the Decision Tree with 85%, Gaussian Naïve Bayes with 82% and Support Vector Machine with 76% respectively.