🔗 Uplift Random Forests

Uplift Random Forests#

An ensemble of trees that estimates individual-level treatment effects.

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 Despite careful review, it may still contain mistakes, omissions, or out-of-date information.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#

An uplift random forest is a modified random forest built to estimate the causal effect of a treatment directly, not merely to predict an outcome. Instead of modelling \(P(Y \mid X)\), it models the treatment-versus-control difference,

\[\Delta(X) = P(Y = 1 \mid T = 1, X) - P(Y = 1 \mid T = 0, X).\]

How it works#

The trees are uplift trees: each split is chosen to maximise the difference in treatment effect between its branches, rather than to maximise class purity. Many such trees are then averaged in an ensemble, exactly as in an ordinary random forest, for stability — and each individual receives an estimated uplift, the incremental probability change caused by the treatment.

Why use it#

It handles nonlinear relationships and feature interactions automatically, reduces variance compared with a single uplift tree, and delivers individual-level treatment-effect predictions — learning how features modify the treatment effect, not just how they drive the outcome.

Applications#

It powers marketing (targeting customers who respond because of a campaign), personalised medicine (patients who benefit most from a drug), and policy (subgroups most positively affected). It requires both treated and control data — an A/B setup. In an email sign-up campaign, a standard forest predicts the probability of signing up, while the uplift forest predicts the extra probability caused by the email — separating loyal always-signers (low uplift) from persuadables (high) and negative reactors (negative uplift).

Theme: Causal Inference & Uplift · All terminology

Hint

Mind map — connected ideas

Uplift Models · Uplift · Treatment Effect · Causal Inference · Uplift Score · Causal ML (Causal Machine Learning)

Hint

More in Causal Inference & Uplift

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