Interpretable machine learning for high-dimensional trajectories of aging health
Fig 4
Heatmap of the posterior mean value of the robust network weights. Weight directions are read from the horizontal axis (j) towards the vertical (i), Wi←j. The sign and color of the weight signify the direction of effect—a positive weight implies that an increase in a variable along the horizontal axis influences the vertical axis variable to increase. A negative weight implies that an increase in a variable along the horizontal axis influences the vertical axis variable to decrease. Hierarchical clustering is applied to the absolute posterior mean of the robust weights to create a dendrogram (at right).