An Effective Machine Learning Techniques to Detect Parkinson's Disease

Abstract

Parkinson’s disease is one type of neurological disorders that affects the systema nervosum and causes unintended or uncontrollable movement in the body parts. More than 6 million people all over the world were affected by PD disease. It is difficult to identify the disease at its early stages. Signs of the disease may be can vary from person to person. Symptoms usually begin with a tremor in one hand and gradually start affecting the whole body. At present, there is no clinical equipment or process to recognize this disease at the beginning stage of Parkinson’s disease. Doctors usually diagnose the person by taking a previous medical history and MRI images of the person’s brain and also by observing the symptoms of the person manually which takes more time and cannot detect the disease at its early stages. This disease can be detected at early stages using a machine learning approach with high accuracy. Voice and spiral drawing dataset are collected from normal and PD-affected people and is given as input. 60% of the total dataset is used to train and build the model and the resting 40% dataset is used to test the model. By applying Linear regression and support vector machine and KNN algorithms on voice data sets, this system measures the deflections in the voice of a person. Accuracy with different algorithms is measured. Random forest and CNN algorithms are applied to the spiral data set. Random forest converts spiral drawings into pixels which are very helpful for classification. At the time of testing, the pixels of the current drawing are compared with the previously trained models to detect the disease. By combining the results of the voice dataset and spiral drawings dataset, the machine will detect the disease with high accuracy. The data of a person can be entered into the dataset to detect the disease.