Skip to main content

Table 3 Explanation design preferences

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

Desired content (what)

Benefits

Preferred design

(how)

Explanations

Help assess model credibility and utility

Risk expressed as percent probability

High-level information with details available on demand

Interactive options to support different displays/organizations for various users

Table of raw feature values

Interpret discretized features

Examine trend-based features

Directionality for trend-based features

Simpler terminology

Time-series data plot

Investigate suspicious values

Assess trends and baselines

Multiple plots

Highlight points related to features

Auto-population of data

Contextual information

Clinically meaningful interpretation

Context for risk prediction

Providing clinical context information

Prominent display of baseline risk

Inclusion of risk trends