Acoustic, Prosodic, and Phonological Features for Alzheimer's Detection

M. Zakaria Kurdi's research paper, published on figshare in April 2026, investigates speech-based biomarkers for Alzheimer's disease. The study analyzes acoustic, prosodic, and phonological features from the ADReSS Challenge dataset and the Delaware corpus, achieving an F1-score of 0.89 with ensemble learning. The 627.3 KB PDF details the methodology, feature ranking, and model performance.

Use Cases

• Training machine learning models for Alzheimer's detection based on acoustic features.
• Analyzing the complementary role of prosodic and phonological features in speech for cognitive assessment.
• Benchmarking new models against reported results on the ADReSS Challenge dataset.
• Exploring lexicon-independent speech analysis for early disease diagnosis.

Strengths

• Achieved a reported F1-score of 0.89 for Alzheimer's detection using ensemble learning.
• Systematically evaluated four feature ranking methods and seven machine learning models.
• Validated findings on multiple datasets, including the ADReSS Challenge and Delaware corpus.

Limitations

• The dataset is a 627.3 KB PDF; the underlying feature data is not directly provided.
• Column-level documentation and sample data are unavailable, requiring inference from the paper.
• Row count and specific feature values are unknown, limiting direct usability for modeling.

Provenance

Source

M. Zakaria Kurdi via figshare

Collection Method

Analysis of the publicly available ADReSS Challenge dataset and the Delaware corpus.

Time Range

The study uses existing speech datasets; specific recording dates are not provided.

Freshness

Last updated 2026-04-16 05:24:09

Geography

Geographic coverage of the source datasets is not specified.