Abstract:
Parkinson's disease is a nervous system disorder generally found in the elderly. A common symptom exhibiting in Parkinson’s patients is a deterioration of handwriting caused by slowing of movement and feelings of muscle stiffness in hands and fingers of the patients. Recently, handwriting has been used for clinical assessment based on standard scoring by medical professionals. However, this method requires expertise and is labor intensive. Therefore, machine learning is used to analyze handwriting signals. In our project, we use the database called “the Parkinson's Disease Handwriting Database (PaHaW)” provided by the BDALab. It contains handwriting datasets of healthy and
Parkinson's patient. The handwriting signals were quantified by using feature extraction methods such as such as kinematic analysis, entropy, pressure analysis, power spectral density. To select the relevant features Mann-Whitney U test mRMR and relief algorithm were applied prior to the decision tree classification. The proposed classification system can identify Parkinson's patients from healthy participants with an accuracy of 86.67%.
