Performance comparison of ML methods applied to motion sensory information for identification of vestibular system disorders

This study is the first step gone to develop a Machine Learning (ML) algorithm to be applied to sensory information collected from people to identify Vestibular System (VS) disorders. Three ML methods, the Support Vector Machine (SVM), SVM with Gaussian Kernel and Decision Tree are compared to determine the one with the highest accuracy to use for VS analysis. These methods are applied to the data set collected from groups both of healthy and suffering from VS disorders. All three methods had computation time in tens of milliseconds providing the possibility of real time processing in the field of identification of diseases related to VS imperfections. The assessment of the algorithms was based on processing of 22 features extracted from the dataset. SVM with Gaussian Kernel performed best with 81.3% accuracy. Following this step, some addition and removal of features is made to observe their effect on the training model. We noticed that some features are discriminative that they have significant influence on the overall accuracy. Thus, as a next step, the objective of this work is to apply some feature selection methods to find the most discriminative features to decrease the algorithm complexity while increasing the system accuracy. The ultimate goal of our study is to develop a ML algorithm embedded in wearable devices in order to diagnose people with VS-problems in their daily life.