Constrained optimization-based extreme learning machines with bagging for freezing of gait detection

Abstract

The Internet-of-Things (IoT) is a paradigm shift from slow and manual approaches to fast and automated systems. It has been deployed for various use-cases and applications in recent times. There are many aspects of IoT that can be used for the assistance of elderly individuals. In this paper, we detect the presence or absence of freezing of gait in patients suffering from Parkinson’s disease (PD) by using the data from body-mounted acceleration sensors placed on the legs and hips of the patients. For accurate detection and estimation, constrained optimization-based extreme learning machines (C-ELM) have been utilized. Moreover, in order to enhance the accuracy even further, C-ELM with bagging (C-ELMBG) has been proposed, which uses the characteristics of least squares support vector machines. The experiments have been carried out on the publicly available Daphnet freezing of gait dataset to verify the feasibility of C-ELM and C-ELMBG. The simulation results show an accuracy above 90% for both methods. A detailed comparison with other state-of-the-art statistical learning algorithms such as linear discriminate analysis, classification and regression trees, random forest and state vector machines is also presented where C-ELM and C-ELMBG show better performance in all aspects, including accuracy, sensitivity, and specificity.