Fig. 1

General workflow of the proposed method. Our strategy relies on a topological analysis to perform dimension reduction of both the training (light green) and test data (dark green). Data matrices are transformed into graphs, from which topological features are extracted. Even if the original features (light blues) are different, the topological features extracted from the graphs (dark blue) have the same meaning and are comparable. These features are then used to train and test several models that rely on different learning algorithms (DNN, SVM and RF). These models are compared based on the accuracy of their predictions on the test data