Feature Selection of Arabic Online Handwriting Using Recursive Feature Elimination for Parkinson’s Disease Diagnosis

Abstract

Parkinson’s disease (PD) is one of the most common neurodegenerative diseases affecting a large population worldwide. Parkinson’s disease is characterized by rigidity, slowness of movement and tremors at rest, these syndromes are frequently manifested in the deterioration of handwriting. The aim of this article is to perform online Arabic handwriting analysis for two types of tasks, TASK 1: copying arabic imposed text and TASK 2: writing arabic desired text. A novel method of handwriting selection features is proposed to obtain the relevant features to efficiently identify subjects with PD, based on Recursive Feature Elimination with Cross-Validation (RFECV), three different RFE estimators were compared: Support Vector Machine, Decision Trees and Random Forest, the selected features have been fed to the same classifiers above to determine the best classifier for predicting Parkinson’s disease. Result: An accuracy of 94.4% was obtained using SVM with Linear kernel, based on 55 features selected using RFE-SVM(Linear) for TASK 1, for TASK 2 an accuracy of 93.7% was obtained using SVM with RBF kernel, based only in 7 features selected using RFE-SVM(Linear). For all the classifiers used, this technique experimentally demonstrates an increase in performance metrics.