Fig. 5

Model explanation of the best performing model, RSF trained on time series extracted features, using TIME. TIME (Temporal Importance Model Explanation) is a model-agnostic longitudinal explanation method. A cell (box) is colored if it’s important, is white if not deemed important by the model. Each row is a variable and shows the most important windows to the model (groups of cells in the same shade of color). The variables are ordered along the y-axis based on the overall importance (darker color = more important). Hatched texture implies the ordering within the window is important to model prediction (i.e., shuffling the variable values at different times within the window affects the model prediction)