Skip to main content
Fig. 6 | BMC Medical Informatics and Decision Making

Fig. 6

From: A qualitative research framework for the design of user-centered displays of explanations for machine learning model predictions in healthcare

Fig. 6

Final user-centered explanation display. The predicted risk and baseline risk are displayed in percent probability at the top of the figure. The explanation plot (top left) uses feature groups as the unit of explanation, but has hover-box capability to view individual features within each feature group. In the hover-box, trend-based features are summarized by trend direction (e.g., “Cr has increased since minimum value”). The plot includes interactive controls to view additional feature groups (e.g., scrolling down the explanation plot) and view different sets of feature groups (e.g., view laboratory test feature groups). The table of raw feature values (bottom left) includes the description, value, and contribution to the predicted risk for each individual feature. This table also includes the trend direction for trend-based features. The time-series plots to display raw values of laboratory test and vital sign data (right) highlight the points used to compute features and include interactive controls to zoom in on and select regions of data. These plots also have hover functionality that can be used to show the value and time of a specific point. To facilitate data exploration, interactivity is linked across plots and tables (e.g., selecting a predictor on the explanation plot will highlight it in the raw feature table and load the appropriate laboratory test/vital sign in the time-series plot)

Back to article page