Causal Exploring Analysis Project
This project demonstrates the use of causal exploring algorithms applied to lifestyle and medical data. Our goal is to identify underlying structures that influence health outcomes.
Understanding Causal Relationships
Causal exploring goes beyond simple correlation. While correlation tells you that two variables are linked, causality tells you which one influences the other and how. In complex lifestyle datasets, understanding these direct mechanisms is essential to draw meaningful conclusions from data.
Our approach uses directed acyclic graphs (DAGs) to represent causal relationships between variables, inferred directly from observational data without requiring prior knowledge of the relationships.
Methodology
Our research pipeline applies three causal discovery algorithms:
- PC-Stable (R Package: bnlearn): A constraint-based algorithm that tests conditional independence between variables to infer the causal skeleton, then orients edges where possible. Stable version ensures consistent results regardless of variable ordering.
- MRPC (R Package: MRPC): An extension of PC that incorporates prior knowledge constraints. Particularly suited for datasets where some variables are known to be upstream causes.
- MMHC (R Package: bnlearn): A hybrid algorithm combining a constraint-based phase (MMPC) to identify candidate neighbors, followed by a score-based phase (Hill Climbing) to orient the edges. More robust on noisy datasets.
How to use the tool
Navigate to the Home page, upload your CSV dataset and select the algorithm you want to apply. The resulting causal graph will be displayed on the Results page, where you can explore the inferred relationships between variables.