870. Estimating True Demand Under Stockouts and Interruptions: A Hybrid Tobit Kalman Filter and Conformal Prediction Approach
Invited abstract in session TA-38: Forecasting, prediction and optimization 1, stream Data Science meets Optimization.
Tuesday, 8:30-10:00Room: Michael Sadler LG19
Authors (first author is the speaker)
| 1. | Harsha Halgamuwe Hewage
|
| Data lab for Social Good Research Group, Cardiff Business School, Cardiff University | |
| 2. | Bahman Rostami-Tabar
|
| Data lab for Social Good Research Group, Cardiff Business School, Cardiff University | |
| 3. | Aris Syntetos
|
| Cardiff Business School, Cardiff University |
Abstract
Accurate demand estimation is critical for family planning supply chain management, particularly in contexts where stockouts and operational disruptions frequently lead to censored demand observations. Standard forecasting models often fail to capture true demand due to lost sales. In this study, we propose a hybrid Tobit Kalman Filter and conformal prediction framework to reconstruct true demand by correcting censored observations. We model demand as a state-space system incorporating baseline level, trend, seasonality, and a dynamic missing demand correction component. The Tobit Kalman Filter estimates true demand by adjusting for stockouts, interruptions and external factors such as high and low consumption indicators. To address uncertainty in censored demand estimation, we integrate conformal prediction with residual sampling, providing distribution-free prediction intervals instead of relying on restrictive normality assumptions. This ensures robust uncertainty quantification, improving true demand estimation even under highly volatile and non-Gaussian demand patterns. Unlike existing models, our approach explicitly differentiates partial censorship (stockouts) from full censorship (interruptions and no stock received), enabling context-sensitive corrections. We evaluate the performance of our proposed model through simulation studies and real-world contraceptive supply chain data, presenting a scalable solution with significant implications for inventory management.
Keywords
- Forecasting
- Health Care
- Supply Chain Management
Status: accepted
Back to the list of papers