Abstract Dementia is a debilitating disease that leads to a gradual loss of memory and other cognitive abilities and tends to develop in people over 60 years of age. The number of people worldwide with dementia is expected to increase from 57.4 million in 2019 to 131.5 million in 2050. Diagnosis of dementia is a critical task, and some biomarkers and psychological and demographic measures have been used to diagnose dementia clinically. The advent of artificial intelligence (AI) has created new opportunities to improve diagnostic accuracy. In this study, we propose an approach based on machine learning and graph analysis to identify important patterns and diagnostic markers for dementia. The novel contributions of this study include a descriptive probability density function (PDF) analysis to examine the distribution of clinically relevant markers across diagnostic groups, providing exploratory insights into group-wise feature behavior. A comprehensive evaluation of ten machine learning (ML) models is conducted under strict subject-level data separation to identify robust classification performance. In addition, feature ranking is assessed using three complementary strategies (Random Forest importance, Chi-Square statistics, and Boruta ranking) to evaluate the consistency and robustness of marker relevance. Out of ten ML models evaluated, the Random Forest achieved the highest accuracy of 94.37%, with a macro F1-score of 86.96%. The result of the ML model is further validated using a graphical analysis of critical markers. The results of this study might help to improve the diagnostic accuracy.
Sutradhar et al. (Mon,) studied this question.