Abstract: An example method includes identifying training data indicating features of a sample population and clinical outcomes of the sample population. The clinical outcomes are associated with a heterogeneous condition. The method further includes generating decision trees in a Random Forest (RF) based on the training data, each one of the decision trees being configured to divide the sample population into multiple categories based on the features of the sample population. In response to generating the decision trees, a proximity matrix comprising multiple entries is generated using the RF. One of the entries indicates a proportion of the decision trees that categorize a first individual among the sample population and a second individual among the sample population into the same categories among the multiple categories. The method further includes identifying subgroups of the heterogeneous condition by detecting communities of the proximity matrix.

Abstract: An example method includes identifying training data indicating features of a sample population and clinical outcomes of the sample population. The clinical outcomes are associated with a heterogeneous condition. The method further includes generating decision trees in a Random Forest (RF) based on the training data, each one of the decision trees being configured to divide the sample population into multiple categories based on the features of the sample population. In response to generating the decision trees, a proximity matrix comprising multiple entries is generated using the RF. One of the entries indicates a proportion of the decision trees that categorize a first individual among the sample population and a second individual among the sample population into the same categories among the multiple categories. The method further includes identifying subgroups of the heterogeneous condition by detecting communities of the proximity matrix.