Predicting Dementia Risk Using Paralinguistic and Memory Test Features with Machine Learning Models

Abstract

Cognitive reserve exposures are a major class of dementia risk predictors, but a biomarker has proven elusive. Here, we show that paralinguistic features extracted from audio recordings of older participants completing the LOGOS episodic memory test can be used to identify participants with high and low estimable cognitive reserve, and hence low and high dementia risk, respectively. We present a parallel classification system consisting of an ensemble of a k-NN model and SVM model that discriminates between participants at high risk and low risk of dementia with an accuracy of 94.7% when trained with paralinguistic features only and 97.2% when trained with paralinguistic and episodic memory features.