Utilising pattern recognition algorithm to determine the possible speech score for Alzheimer's disease prediction

This study employs machine learning (ML) to support the early-stage diagnosis of Alzheimer's disease (AD) using speech score data as a non-invasive, cost-effective alternative to traditional methods, like imaging and biomarker testing. Conventional techniques are often inaccessible, invasive and lack a strong correlation with cognitive decline. Speech-based assessments capture fine cognitive and linguistic deficits, early indicators of AD. The study employs six cognitive parameters: MoCA v7.1, HVLT-R (immediate and delayed recall), BNT-SF, and WMS-R Logical Memory I and II. A classification model integrating random forest, SVM and KNN was evaluated using k-fold cross-validation. ANOVA-based feature selection enhanced model performance, with random forest and SVM achieving high AUC scores of 0.98 and 0.97, respectively, outperforming KNN. Among features, MoCA and HVLT-R scores were most critical in distinguishing mild cognitive impairment (MCI) from healthy individuals. The findings highlight the potential of speech scores as early biomarkers for AD detection and monitoring.