Figure 1.

Supervised Scaling Approach. A null Cox model is trained in order to obtain a proxy dependent variable (1), e.g. martingale residuals. The fitted coefficients obtained from training a supervised learning method, e.g. linear regression, are used to scale our feature space (2). A clustering method is applied over the scaled feature space (3). The clustering implementation here shown is consensus clustering over 1k runs of the k-median (k = 2) clustering method using different initial seeds and Manhattan distance as the dissimilarity measure.