Characterization of long-lived and non-long lived profiles through biclustering

The understanding of the variables that influence the human ageing process can help in the development of public policies that improve the quality of life of the elderly population. In previous experiments on a human ageing dataset, we applied biclustering techniques on the binary data in order to find long-lived and non-long lived profiles, but we only found long-lived profiles. Then in this work, we propose to use Factor Analysis to represent this data in reduced dimensionality, generating three datasets where the variables with high correlation with each other belong to the same factor. We observed that some variables have a high correlation with each other in the three datasets, allowing them to be grouped into blocks of correlated variables. Posteriorly we applied biclustering on these datasets and validate the results using p-Value measure, Jaccard similarity, and the priori knowledge of the classes. In this way, we found factors belonging only to non-long lived biclusters and biclusters representatives of both profiles.