Improving the prediction accuracy of Parkinson’s Disease based on pattern techniques

Abstract

Parkinson’s Disease is a locomotive disorder commonly found among elders and causes various physical prodromes in an affected personnel. Freezing of Gait is a prominent symptom and gait data may be used to identify the occurrence of this event. The proposed study employs binary pattern recognition such as Local Binary Pattern and Extended Local Binary Pattern to transform gait data into a normal distribution and then extract statistical features such as skewness, kurtosis and etc. The procured features were then classified by different classifiers to determine the model with highest performance. The performance metrics evaluated were accuracy, precision and recall, and an accuracy of 98.82% were achieved for the classifier Logistic Regression.