ECML/PKDD’23 Workshop on Uplift Modeling and Causal Machine Learning for Operational Decision Making

Topic

Uplift modeling (UM) and causal machine learning (CML) for operational decision making (ODM) concerns the discovery and estimation of causal effects from data for optimizing, automating or customizing operational decision-making. The field receives a growing interest from both academics and industry practitioners, with applications in marketing, process management, pricing, medicine, machine maintenance, operations management, human resources, etc.

UM & CML entail a diverse range of specialized data-driven methods, drawing from both the fields of causal inference and machine learning. These methods can either learn (1) from experimental data obtained through randomized controlled trials (RCT), which in some settings is commonly available (e.g., A/B test data in marketing), or (2) from observational data, which is gathered by observing ongoing processes and operations subject to the current decision-making policy. Uplift modeling and causal machine learning extend upon predictive modeling (i.e., supervised learning) and involve additional challenges and complexity, for instance, related to addressing selection bias or evaluating counterfactual predictions. The approaches require special methodology to address issues such as the fundamental problem of causal inference (unobservability of counterfactual outcomes). The field differs from other areas of causal discovery by focusing on practical applications and business problems. A large number of open research questions and practical challenges towards adopting uplift modeling and causal machine learning in practice are still to be addressed and the domain would benefit from further formalization.

This workshop aims at bringing together, for the second time at ECML/PKDD, researchers and practitioners working on UM & CML for DM, to present and discuss recent developments, to identify open issues and to form a community and foster future initiatives.

The workshop is a continuation of the ECML/PKDD'22 Uplift Modeling Tutorial and Workshop. The website of the 2022 edition is available here.

Call for papers

We invite original contributions related to uplift modeling and causal machine learning for operational decision-making. Both methodological and application-oriented submissions are welcomed.

Accepted papers will be publised in a Springer volume of ECML/PKDD'23 workshop proceedings.

The list of topics includes but is not limited to:

Novel uplift modeling and causal machine learning techniques
Learning to rank for uplift modeling and causal prediction
Procedures and measures for evaluation
Practical causal discovery and causal effect estimation under biased treatment assignment
Beyond binary treatments and binary outcomes: continuous, multi- and high-dimensional treatments, continuous or complex (multiple, time-dependent, etc.) outcomes
Cost-sensitive uplift modeling and causal machine learning
Practical applications and case studies
Descriptions of datasets and benchmarking experiments

Submission guidelines

Submitions should be made HERE.

Various types of submissions are encouraged: full papers and extended abstracts. We also make it possible to directly submit a poster you intend to present at the workshop.

Papers and abstracts should adhere to the general paper formatting guidelines as provided on the ECML/PKDD2023 conference website in the Paper Format section.

Authors of accepted submissions will be invited to present their paper in an oral presentation (accompanied by a poster) or a poster presentation and will have an option of publishing the paper in Springer post-proceedings volume.

Please note that at least one author of each accepted paper must have a full registration and be in Turin to present the paper.

Keynote Presentation

Gaël Varoquaux, INRIA, France

Title: “Individualizing treatment effects: transportability and model selection”

Abstract: For efficient interventions, we would like to know the causal effect of the intervention on a given individual: the individual treatment effect. Given the proper set of covariates, such quantity can be computed with machine-learning models: contrasting the predicted outcome for the individual with and without the treatment. I will analyse in detail how to best compute such quantities: what choice of covariates to minimize the variance, how to empirically select the best machine-learning model, and how a good choice of population-level summaries of treatment effect is least sensitive to heterogeneity.

About the speaker: Gaël Varoquaux is a research director working on data science at Inria (French Computer Science National research) where he leads the Soda team on computational and statistical methods to understand health and society with data. Varoquaux is an expert in machine learning, with an eye on applications in health and social science. He develops tools to make machine learning easier, suited for real-life, messy data. He co-funded scikit-learn, one of the reference machine-learning toolboxes, and helped build various central tools for data analysis in Python. He currently develops data-intensive approaches for epidemiology and public health, and worked for 10 years on machine learning for brain function and mental health. Varoquaux has a PhD in quantum physics supervised by Alain Aspect and is a graduate from Ecole Normale Superieure, Paris.

Program Committee

Chairs / Ogranizers

Eustache Diemert
Criteo, France

Szymon Jaroszewicz
Polish Academy of Sciences & Warsaw University of Technology, Warsaw, Poland

Wouter Verbeke
KU Leuven, Belgium

Members

Bart Baesens, KU Leuven, Belgium
Jeroen Berrevoets, University of Cambridge, U.K.
Artem Betlei, Criteo, France
Robin Gubela, Humboldt University, Germany
Stefan Feuerriegel, LMU Münich, Germany
Stefan Lessmann, Humboldt University, Germany
Krzysztof Rudaś, Polish Academy of Sciences, Poland
Toon Vanderschueren, KU Leuven, Belgium
Sam Verboven, Vrije Universiteit Brussel, Belgium

Accepted Papers

Individual treatment effect in Stroke with conformal inference
Sermkiat Lolak (Department of Clinical Epidemiology and Biostatistics, Faculty of Medicine, Ramathibodi Hospital, Mahidol University, Bangkok)

Using Uplift Models to Avoid Student Attrition in Universities
Sebastian Maldonado (University of Chile); Jaime Miranda ( University of Chile); Jonathan Vásquez (Universidad de Valparaiso); Diego Olaya (Free University Brussels)

Improving incentive policies to salespeople cross-sells through uplift modeling
Carla Vairetti (Universidad de los Andes); Sebastian Maldonado (University of Chile); Catalina Sánchez (Universidad de los Andes); Guillermo Armelini (Universidad de los Andes); Andrés García (Universidad de los Andes)

Exploiting causal knowledge during CATE estimation using tree based metalearners
Roger Pros Rius (Universitat de Barcelona); Jordi Vitria (Universitat de Barcelona)

A Parameter-Free Bayesian Framework for Uplift Modeling - Application on Telecom Data
Mina Rafla (Orange Labs); Bruno Cremilleux (Université de Caen Normandie); Nicolas Voisine (Orange)

A Causal Perspective on Loan Pricing -- Investigating the Impacts of Selection Bias on Bid Response Modelling
Christopher Bockel-Rickermann (KU Leuven)

The impact of heteroscedasticity on uplift modeling
Björn Bokelmann (Humboldt University Berlin); Stefan Lessmann (Humboldt University of Berlin)

NOFLITE: Learning to Predict Individual Treatment Effect Distributions
Toon Vanderschueren (KU Leuven); Jeroen Berrevoets (University of Cambridge); Wouter Verbeke (KU Leuven)

A churn prediction dataset from the telecom sector: a new benchmark for uplift modeling
Théo Verhelst (Université Libre de Bruxelles); Denis Mercier (Orange Belgium); Jeevan Shrestha (Orange Belgium); Gianluca Bontempi (Université Libre de Bruxelles)

Schedule

The workshop will take place on Friday September 22nd in the afternoon in PoliTo Room 1T.
Schedule of the workshop:

Workshop presentations Part I
14:30--14:40	Opening remarks Szymon Jaroszewicz (Polish Academy of Sciences); Wouter Verbeke (KU Leuven); Eustache Diemert (Criteo)
14:40--14:55	A churn prediction dataset from the telecom sector: a new benchmark for uplift modeling Théo Verhelst (Université Libre de Bruxelles); Denis Mercier (Orange Belgium); Jeevan Shrestha (Orange Belgium); Gianluca Bontempi (Université Libre de Bruxelles)
14:55--15:10	The impact of heteroscedasticity on uplift modeling Björn Bokelmann (Humboldt University Berlin); Stefan Lessmann (Humboldt University of Berlin)
15:10--15:25	NOFLITE: Learning to Predict Individual Treatment Effect Distributions Toon Vanderschueren (KU Leuven); Jeroen Berrevoets (University of Cambridge); Wouter Verbeke (KU Leuven)
15:25--15:40	Exploiting causal knowledge during CATE estimation using tree based metalearners Roger Pros Rius (Universitat de Barcelona); Jordi Vitria (Universitat de Barcelona)
15:40--15:55	A Parameter-Free Bayesian Framework for Uplift Modeling - Application on Telecom Data Mina Rafla (Orange Labs); Bruno Cremilleux (Université de Caen Normandie); Nicolas Voisine (Orange)
16:00--16:30	Coffee break
16:30--17:10	Keynote by Gaël Varoquaux, INRIA, France Individualizing treatment effects: transportability and model selection
Workshop presentations Part II
17:10--17:25	Using Uplift Models to Avoid Student Attrition in Universities Sebastian Maldonado (University of Chile); Jaime Miranda ( University of Chile); Jonathan Vásquez (Universidad de Valparaiso); Diego Olaya (Free University Brussels)
17:25--17:40	A Causal Perspective on Loan Pricing -- Investigating the Impacts of Selection Bias on Bid Response Modelling Christopher Bockel-Rickermann (KU Leuven)
17:40--17:55	Individual treatment effect in Stroke with conformal inference Sermkiat Lolak (Department of Clinical Epidemiology and Biostatistics, Faculty of Medicine, Ramathibodi Hospital, Mahidol University, Bangkok)
17:55--18:10	Improving incentive policies to salespeople cross-sells through uplift modeling Carla Vairetti (Universidad de los Andes); Sebastian Maldonado (University of Chile); Catalina Sánchez (Universidad de los Andes); Guillermo Armelini (Universidad de los Andes); Andrés García (Universidad de los Andes)

Topic

Call for papers

Submission guidelines

Key dates