Automatic Classification of Parkinson’s Disease Based on Severity Estimation

Abstract

The analysis and classification of Parkinson’s disease is of high clinical importance. The simple and non-invasive method involving gait parameters is attractive to clinicians as well as researchers. This study presents an approach for the identification of Parkinson’s disease depending on the gait parameters. The proposed model estimates disease severity with Gradient Boosted tree regression method. A large number of artificially constructed features based on gait parameters were used in the model. It gives the mean absolute error of 0.5 on the Unified Parkinson’s disease rating scale (UPDRS). The present work also demonstrates that the generated set of features is important in estimating the severity of Parkinson’s patients.