Criteo Uplift Balanced: 25M User Records for Advertising Incrementality Tests

25,309,483 anonymized user records assembled from randomized advertising incrementality tests by Criteo AI Lab. Each row includes 11 anonymized features, a treatment indicator, and binary labels for visits and conversions. The dataset has been preprocessed and balanced for classification and clustering tasks.

Use Cases

• Benchmarking uplift modeling algorithms based on treatment and outcome labels.
• Analyzing heterogeneity of treatment effects using the anonymized feature set.
• Developing methods for observational causality based on the assembled incrementality test data.
• Training classification models to predict user visits or conversions from anonymized features.

Strengths

• Large scale with 25,309,483 rows providing statistical power.
• Includes a treatment indicator and two outcome labels (visit and conversion) specifically for causal tasks.
• Preprocessed and balanced for classification, with label proportions detailed in the description.

Limitations

• All 11 features are anonymized and randomly projected, making semantic interpretation impossible.
• The dataset is a non-uniform subsample for privacy, which may alter original incrementality levels.
• Column-level documentation for the specific adapted version is absent; the original schema must be inferred.

Provenance

Source

Criteo AI Lab

Collection Method

Assembled from several randomized advertising incrementality tests.

Time Range

null

Freshness

Last updated date is unknown; freshness unverified.

Geography

null