Machine Learning Application to Quantify the Tremor Level for Parkinson’s Disease Patients

Abstract

Parkinson's disease (PD) is a neurodegenerative disorder with progressive nature. It causes motor symptoms such as resting tremor, bradykinesia and others movement disorders. Because of its progressive nature, this disease needs a continuous monitoring of motor symptoms. Health Monitoring Systems are widely used to monitor the disease progress, improving the treatment and minimizing the drug side effects. In this research, we developed two predictive models using a supervised machine learning approach. These models can classify the Parkinson disease's rest tremor between high or low frequencies, showing the intensity of Parkinson's motor symptom. This classification allows the detailed monitoring of the medication's effectiveness and the disease progress. In our validation, we applied leave-one-out cross-validation methods to classify the level of the PD tremor. In our results, we reached a classification accuracy of 92.8%. Therefore, this work proposes a new approach to classifying PD tremor and improving the the patients quality´s live on treatment, using non-invasive health monitoring systems improved by machine learning classification algorithms.