Sentence Writing Test for Parkinson Disease Modeling: Comparing Predictive Ability of Classifiers

摘要

The present paper is devoted to the modeling of the sentence writing test to support diagnostics of Parkinson's disease. Combination of the digitalized fine motor tests and machine learning based analysis frequently lead the results of very high accuracy. Nevertheless, in many cases, such results do not allow proper interpretation and are not fully understood by a human practitioner. One of the distinctive properties of the proposed approach is that the set of features consists of parameters that may be easily interpreted. Features that represent size, kinematics, duration and fluency of writing are calculated for each individual letter. Furthermore, proposed approach is language agnostic and may be used for any language based either on Latin or Cyrillic alphabets. Finally, the feature set describing the test results contains the parameters showing the amount and smoothness of the fine motions which in turn allows to precisely pin down rigidity and unpurposeful motions.