🔗 Causal ML (Causal Machine Learning)

Causal ML (Causal Machine Learning)#

ML methods that estimate causal effects rather than predictive associations.

Important

✨ AI-generated content. This page was written with the assistance of an AI language model and is provided as a learning aid. Despite careful review, it may still contain mistakes, omissions, or out-of-date information. Whether you are new to the topic, a team lead, or a senior practitioner, treat it as a starting point rather than an authoritative reference: read it critically and independently verify anything you act on (code, commands, figures, and factual claims) against official documentation and primary sources before relying on it.

What it is#

Causal ML (causal machine learning) is the branch of machine learning that estimates cause-and-effect relationships, not merely correlations or predictions. Where traditional ML asks “what is likely to happen?”, causal ML asks “what will happen because of this intervention?” — making it essential for policy evaluation, medical treatment, marketing and any decision where an action changes the outcome.

The counterfactual problem#

Everything rests on treatment versus control — the intervention against no intervention. The difficulty is the counterfactual problem: for any individual we observe only one outcome (treated or not), never the other, so the effect must be estimated. That effect comes in three grains: the ATE (average treatment effect across the population, \(\text{ATE} = \mathbb{E}[Y(1) - Y(0)]\)), the CATE (conditional on a subgroup), and the ITE (for a single individual).

Methods#

The gold standard is experimental — a randomised controlled trial (RCT) assigns treatment at random, eliminating confounding. When experiments are impossible, observational methods step in: propensity-score matching, inverse-propensity weighting, and doubly-robust estimators that combine regression with weighting. ML extensions add meta-learners (S-, T-, X- and R-learners), causal trees and forests for heterogeneous effects, and deep models (Dragonnet, TARNet) — with libraries such as CausalML, EconML and DoWhy.

An example#

In a marketing email campaign, traditional ML predicts who will buy; causal ML predicts who will buy because of the email. If the treatment group buys at 15% and the control at 10%, the ATE is a 5% uplift — and causal ML then estimates CATE/ITE to reveal which segments or individuals respond most.

Benefits and challenges#

The payoff is real: it captures true causal effect rather than correlation, targets only those who benefit, and supports counterfactual reasoning (“what if?”). The costs are structural — it needs a treatment-control design (experiments or strong assumptions), is sensitive to confounding in observational data, and is harder to explain and validate than standard ML.

Theme: Causal Inference & Uplift · All terminology

Hint

Mind map — connected ideas

Treatment Effect · Causal Inference · Uplift Models · Uplift · Random Targeting Strategy · Conversion Rate Uplift

Hint

More in Causal Inference & Uplift

AUUC (Area Under the Uplift Curve) · Causal Effect · Causal Impact · Causal Inference · Causal Trees · Cumulative Incremental Gain (CIG) · Cumulative Uplift · Incremental Conversions · Incremental Gain · Incremental Recovery Rate (IRR) · Incremental Revenue · Incremental Sales · Qini Coefficient · Qini Curve