Stroke Classification from Facial Features Using Machine Learning

Face represents the important features for the visual assessment of the stroke patient. The facial stroke is known as the inability to move a part, one side, or both sides of the facial muscle from nerve damage. However, stroke and bell's palsy are similar symptoms on the patient's face that are difficult to diagnose. Therefore, this research will propose an efficient machine learning algorithm for facial stroke classification. Our research gathers the imaging data from the two published dataset that consist of the facial stroke images from facial droop and facial paralysis and the facial normal images from facial expression. This dataset selects the smile expression to extract the face features by using the facial mapping landmark method. These landmarks are used to calculate the facial features for training the machine learning algorithms that consist of the support vector machine, k-nearest neighbor, decision tree, random forest, logistic regression, AdaBoost, and Bayesian Classifier methods. These methods were trained by using the training accuracy and evaluated by using the accuracy, precision, recall, and F1-score metrics. From the experimental results, the random forest represents the most accurate percentage of training and testing measurements. Therefore, this research proposes the random forest algorithm to be the efficient algorithm to classify the facial stroke from face features of smile expression.