[Online]FedKPI: Federated Data Quality Monitoring of Business KPIs Across Retail Stores

FedKPI: Federated Data Quality Monitoring of Business KPIs Across Retail Stores
ID:3 Submission ID:218 View Protection:ATTENDEE Updated Time:2026-10-04 23:09:33 Hits:151 Online

Start Time:2026-10-12 15:00 (Asia/Ho_Chi_Minh)

Duration:15min

Session:[S5] Track 5: Emerging Trends of AI/ML » [S5-3] Track 5: Emerging Trends of AI/ML

Video No Permission Presentation File

Tips: The file permissions under this presentation are only for participants. You have not logged in yet and cannot view it temporarily.

Abstract
A retail chain computes the same key performance indicators (KPIs) at every store, but the data feeding those KPIs is produced by per-store pipelines that fail silently and indepen-dently. The chain needs to know, per store, when a reported KPI is wrong because the data broke rather than because the business moved, yet pooling raw transactions into one warehouse is often disallowed for privacy, data residency, or competitive reasons. We propose FedKPI, a federated system in which each store detects distorted KPI periods on its own data and a coordinator classifies the scope of an anomaly as a site-local defect, a chain-wide defect, or a chain-wide genuine business movement, sharing only model parameters and compact statistics. We formulate the problem, give a three-way scope taxonomy that only a cross-site view can resolve, and build an evaluation protocol with computed rather than annotated ground truth by injecting typed defects and business events into the public Rossmann Store Sales data (1,115 stores). Clustered federation matches a centralized upper bound (detection F1 = 0.51) while a single global model trails on heterogeneous store formats (macro F1 0.47 → 0.50); on newly opened, data-poor stores federation lifts detection F1 from
0.33 (local only) to 0.49. The coordinator classifies scope at 0.98 accuracy and never mislabels a genuine chain-wide movement as a chain-wide defect, whereas a store acting alone misreads every movement. Under differential privacy, utility is essentially free for ε ≳ 3 and degrades gracefully thereafter, with minority store formats the first to suffer.
Keywords
Index Terms—federated learning, data quality, KPI monitoring, anomaly detection, differential privacy, retail analytics
Speaker
Jayanth Reddy
Technical stafe independent

Submission Author
Jayanth Reddy independent
Comment submit
Verification code Change another
All comments

CONTACT US

Email: asiancomnet2026@usssociety.org

Website & IT Support: hi@aconf.org

Registration Submit Paper