A three-stage fuzzy classifier method for Parkinson's disease diagnosis using dynamic handwriting analysis

Finding low-cost and insightful methods to reinforce the diagnosis of Parkinson's disease is a major challenge today, and using dynamic handwritten data may be one of the solutions. Artificial intelligence and machine learning methods can be used to create predictive models for disease diagnosis. In the case of machine learning models, in addition to maximising their accuracy, it is important to ensure that prediction models for disease diagnosis explain prediction results. The fuzzy classifiers using a set of IF-THEN statements with fuzzy terms often used in natural discourse suit this purpose. This study proposes a fuzzy classifier method consisting of three-stage: generating the structure, selecting the informative features, and optimising the parameters. Thirty-two variants of the method with different metaheuristic algorithms are recommended. Experiments have been conducted on publicly available handwritten signal databases for diagnosing Parkinson's disease such as ParkinsonHW, PaHaW, NewHandPD. ParkinsonHW contains handwriting of 40 people including 25 patients with Parkinson's disease. Handwriting tasks were drawing spirals and meanders. PaHaW contains handwriting of 75 people including 37 patients with Parkinson's disease. Handwritten tasks consisted of drawing spirals and writing letters, bigrams, trigrams, words and sentences in Czech. NewHandPD contains handwriting of 66 people including 31 patients with Parkinson's disease. Handwriting tasks were drawing spirals, meanders and circles. Results of statistical comparison of efficiency with alternative interpretable classifiers - decision trees and fuzzy genetic systems have shown the advantage of some variants of realisation in the accuracy of the prediction and interpretability. Achieved results may indicate the applicability of the proposed method for constructing fuzzy classifiers as a diagnostic tool.