DEMENTIA RISK ASSESSMENT USING MACHINE LEARNING AND PART-OF-SPEECH TAGS

摘要

© 2023 Little Lion Scientific.Dementia, a set of cognitive decline syndromes distinct from typical age-related degeneration, poses a significant public health challenge. The key to dementia detection lies in analyzing sentence structure and conversational style, particularly in speech. This study focuses on creating and evaluating a machine learning model for non-invasive early dementia detection through speech parameter analysis in everyday conversation. Leveraging the DementiaBank dataset, comprising over 500 voice transcripts from individuals aged 60 and older, the study employs 63 tagged Part-of-Speech (PoS) parameters extracted from chat transcripts. Data from 244 control subjects and 306 dementia patients are used. Machine learning methods, including Random Forest, Deep Neural Network, and Support Vector Machine, achieve respective accuracy rates of 83, 92, and 84. These results underscore the effectiveness of informatics-based machine learning in non-invasive dementia detection using PoS tags. Additionally, the study provides insights into the relative importance of each PoS tag in dementia detection. This research contributes to the growing informatics field of dementia detection and supports the development of less intrusive diagnostic tools.