🔗  Uplift@k

Uplift@k#

The incremental gain captured within the top-k targeted population.

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 Despite careful review, it may still contain mistakes, omissions, or out-of-date information.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

Uplift@k is a performance metric in uplift modelling and causal ML. It measures the incremental effect achieved if you target only the top k% of customers, ranked by the model’s predicted uplift score. In plain terms: if I contact only the top k%, how much extra impact do I get compared to not contacting them?

The formula#

It is the difference in average outcome between the treatment and control groups within the top-k% segment,

\[\text{Uplift@}k = \bar{y}_{\text{treatment}}^{(k)} - \bar{y}_{\text{control}}^{(k)},\]

where the averages are taken over the top k% by predicted uplift. Random targeting yields a small or zero value (treatment and control behave alike); a good model makes treatment clearly outperform control in that segment.

A worked example#

With 10,000 customers, targeting the top 20% (k = 20%) selects 2,000. If, within that segment, the treatment group’s purchase probability exceeds the control group’s by 5 percentage points, then uplift@k = +5pp — the extra impact the model captures by choosing those 2,000.

Uses, and versus uplift#

In marketing it estimates incremental sales from promoting only the top k%; in healthcare, the incremental recovery from treating the top-k patients; in recommendation, the extra engagement from targeting the top-k users. The distinction from plain uplift is subtle but important: uplift asks how effective is the treatment overall?, while uplift@k asks how good is my model at selecting the best subset to treat?


Theme: Causal Inference & Uplift  ·  All terminology



See also

Source article Adapted (context, re-expressed) in our own words from: Uplift@k (insightful-data-lab.com).

Tags: purpose: reference topic: terminology level: advanced