The Visual Causality Analyst: An Interactive Interface for Causal Reasoning.

Area of Science:

• Data Science
• Causal Inference
• Human-Computer Interaction

Background:

• Discovering causal relationships in multivariate data is a key challenge in data analytics.
• Existing causal discovery algorithms may infer spurious correlations, necessitating expert validation.
• Integrating domain expertise is crucial for refining causal networks and ensuring accuracy.

Purpose of the Study:

• To present the Visual Causal Analyst, a novel framework for visual causal reasoning.
• To enable collaboration between domain experts and causal discovery algorithms.
• To improve the accuracy and reliability of inferred causal networks.

Main Methods:

• Developed a visual causal reasoning framework integrating a 2D graph view and statistical parameter displays.
• Designed an interface for users to apply expertise, verify, and edit causal links.
• Incorporated a unified model capable of handling both numerical and categorical variables.

Main Results:

• The Visual Causal Analyst facilitates expert-driven refinement of causal discovery.
• The framework provides interactive tools for understanding complex causal structures.
• Demonstrated effectiveness through case studies on practical datasets.

Conclusions:

• The Visual Causal Analyst enhances causal discovery by incorporating domain expertise.
• The framework offers a unified approach for mixed-type data (numerical and categorical).
• This visual approach improves the identification of valid causal networks from data.