Assessing the Interplay of Attributes in Dementia Prediction Through the Integration of Graph Embeddings and Unsupervised Learning

Explainable Artificial Intelligence (XAI) stands as an evolving domain within modern AI methodologies. It centers on the exploration of relationships between variables, dimensionality reduction (directly associated with importance), and the explanation of model decisions. This paper introduces a methodology rooted in Graph Embeddings and unsupervised learning, specifically applied to the prediction of dementia. The focus is on elucidating the significance of prevalent clinical tests utilized in dementia diagnosis, discerning the achievable accuracy of classification models without incorporating these variable groups, and assessing the consequential impact in terms of medical cost and time efficiency.