# Feature selection in causal discovery

**URL:** <https://discourse.datamethods.org/t/feature-selection-in-causal-discovery/28735>\
**Category:** causal inference\
**Created:** [May 6, 2026, 3:53pm UTC](https://discourse.datamethods.org/t/feature-selection-in-causal-discovery/28735 "2026-05-06T15:53:56Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![AriRoane](https://discourse.datamethods.org/user_avatar/discourse.datamethods.org/ariroane/32/5775_2.png) [@AriRoane](https://discourse.datamethods.org/u/AriRoane)\
**Post date:** [May 6, 2026, 3:53pm UTC](https://discourse.datamethods.org/t/feature-selection-in-causal-discovery/28735/1 "2026-05-06T15:53:56Z")

</div>

Hi! I am using a DAG approach in trying to better specify a research question examining stone-free rate after kidney stone treatment. I have constructed an a priori model using review of relevant literature, but wondering if it would be valuable to run some data-driven algorithms on variables in the dataset to understand if this a priori model aligns with the dataset. I know there are both constraint and non-constraint based methods, but wondering if there is a more systematic framework I should be using for feature definition and selection. One big limitation is that many of the actual mediators/mechanisms in the DAG aren’t measured in the dataset, although many of the parents to those mechanisms are.