Feature Evaluation in Handwriting Analysis for Alzheimer’s Disease Using Bayesian Network

Abstract

Alzheimer’s disease, recognized as the most widespread neurodegenerative disorder worldwide, is intricately linked to cognitive impairments. The cognitive impairments, range from mild to severe and are a risk factor for Alzheimer’s disease. They have profound implications for individuals, even as they maintain some level of daily functionality. In previous studies, it was proposed a protocol involving handwriting tasks as a potential diagnostic tool, and a comparative analysis of well-known and widely-used feature selection approach to determine the most effective features for predicting the symptoms related to cognitive impairments via handwriting analysis. In the presented study, we use a Bayesian Network to conduct further analysis of the most effective features extracted from handwriting. Our objective is to exploit the structural learning of Bayesian Networks to discover correlations among the most effective features for predicting impairment symptoms through handwriting analysis and deepen our understanding of the underlying cognitive functions affected. The results showed that the Bayesian Network chooses features conditionally dependent on the determination of the disease, and several features are selected more times than others, underlining their importance in the diagnosis. Moreover, comparing the results with those achieved by well-known and widely-used feature selection and classification approaches, the Bayesian networks exhibit the best performance by using a reduced set of features.