Levels of disability in the older population of England: An unsupervised and self-supervised learning approach

The English Longitudinal Study of Ageing (ELSA) is a survey that targets the older population of England. In this study, we consider this survey and questions related to socio-demographic, diseases, and lifestyle factors to find an optimal level of disability, and significant factors that affect disability in older ages. One problem with ELSA is that there is no indication of the severity of disability and therefore our target is unknown. We group participants based on positive responses to disability-related questions and compare our results with clusters created by the K-modes algorithm. We evaluate our clusters visually with t-distributed Stochastic Neighbour Embedding (t-SNE), Principle Component Analysis (PCA) and Uniform Manifold Approximation and Projection (UMAP) and quantitatively with cohesion and separation measures. We show that pretraining can help with clusters’ visualisation and clusters created by K-modes are of higher quality than our initial guess.