Two-level SVM model with language markers for (early) detection of Alzheimer's Disease
Abstract: This study presents a novel two-level SVM (Support Vector Machine) model for the automatic (early) detection of Alzheimer’s Disease (AD) using language markers that are independent of lexical semantics. We avoid lexical semantic features because they are subject to high individual variation, thus limiting their predictive power for unseen data. Instead, we focus on morphosyntactic, syntactic, and sentence-level features, which are more stable and potentially allow for easier generalization of the model to other datasets, languages, and individuals. We constructed SVMs at both the sentence level and the subject level, applying language features extracted from automatically parsed transcriptions from the Pitt and Delaware corpora in DementiaBank. Our model demonstrated that the subject-level SVM significantly improved classification accuracy. The model yields high performance across all evaluation metrics on the test set for both AD and Mild Cognitive Impairment statuses.
Paper Type: Long
Research Area: NLP Applications
Research Area Keywords: Language Modeling, NLP Applications
Contribution Types: NLP engineering experiment, Data analysis
Languages Studied: English
Submission Number: 4046
Loading