Speech Features for Monitoring Alzheimer’s Disease Using Random Forest Classifier

Detecting Alzheimer’s disease (AD) and disease progression based on the patient's speech data can aid non-invasive, cost-effective, real-time early diagnostic and repetitive monitoring in minimum time and effort using machine learning (ML) classification approaches. The paper aims to predict early AD diagnosis and evaluate stages of AD through exploratory analysis of acoustic features, non-stationarity, and non-linearity testing, and applying data augmentation techniques on AD and cognitively normal (CN) spontaneous speech. After evaluating the proposed AD and AD stages classification models, an accuracy of 82.2% and 71.5% was achieved using the Random Forest classifier, respectively. This will enrich the Alzheimer’s research community with further understanding of methods to improve models for AD classification and addressing nonstationarity and non-linearity principles on audio features to determine the best-suited acoustic features for AD identification.