Apologies for cross-posting.
Dear Colleagues and Students,
INFORMS Service Science Online Forum Series - Episode 11 features our next speakers Profs. Nan Kong (Purdue University) and Haiyan Yu (First Affiliated Hospital of Chongqing Medical University & Chongqing University of Posts and Telecommunications), who will present their research on delivering the right service to the right subgroup under severe data scarcity, with important implications for healthcare operations, personalized service, and equitable decision-making.
See you online for Episode 11 via this Zoom link, where healthcare analytics meets service science!
Speaker: Profs. Nan Kong (Purdue University) and Haiyan Yu (First Affiliated Hospital of Chongqing Medical University & Chongqing University of Posts and Telecommunications)
Moderator: Prof. Muge Yayla-Kullu, UT San Antonio
Topic: Stratified Service under Partial Observation Scarcity
Abstract: Stratified service asks a fundamental question: how can we deliver the right service to the right subgroup when data are abundant for some populations but extremely scarce for others? This challenge arises across many domains, including stratified treatment, targeted marketing, customized supply chain design, and resource allocation for specialized populations such as children and adolescents in clinical trials and public service studies. Although stratification enables more effective and personalized service provision, it also introduces a critical and often overlooked problem: partial observation scarcity. That is, within a given stratum, e.g., patients with the same disease stage or customers with similar demographic profiles, some subgroups are severely underrepresented in the observational data. This imbalance amplifies the risks of biased estimation, unstable learning, and inequitable service policies.
To address this challenge, we propose a weighting-estimation-then-optimization (WETO) framework for stratified service analytics under uneven data availability. The framework combines two key ideas. First, we leverage subgroup pattern transfer (SP) to borrow statistical strength from well-represented subgroups and improve inference for data-poor populations. Second, we incorporate entropy balancing to mitigate confounding effects from individual-level covariates within each subgroup, thereby improving comparability and robustness.
Using real-world clinical records from schizophrenia patients, we demonstrate that a boosting-enhanced version of WETO substantially reduces PANSS scores across all patient strata and consistently outperforms conventional approaches. Importantly, the boosting framework also improves imputation accuracy for underrepresented subgroups relative to standard regression-based methods, highlighting its ability to support equitable decision-making under sparse observations.
Beyond healthcare, the proposed framework generalizes naturally to a broad class of service systems that require personalized decisions under heterogeneous and imbalanced data environments. Potential applications include targeted promotions for niche customer segments, customized product-service strategies for low-volume supply chain partners, and adaptive resource assignment in pediatric and other small-population clinical settings.
Time: May 27, 2026, Wednesday, 11:00 AM-12:00 PM Central Time
Zoom link: https://utsa.zoom.us/j/96054445117
Add to Google Calendar
For more information, see our website https://sites.google.com/view/service-science-online-forum/, and our YouTube Channel https://www.youtube.com/playlist?list=PLCn8oCTLj5JEeIiA3_ATZp8gtlkWJCRpO. Join our mail list to get the information about new episodes of Service Science Online Forum: https://forms.gle/k5L52JZbW8kpLYrf7.
------------------------------
Renyu Zhang
Associate Professor
The Chinese University of Hong Kong
Hong Kong
------------------------------