EUVIMEDEUVIMED entdecken
Crossref

When Temporal Features Reuse Data: Adjacent-Window Bias in Customer Panels

Dr. Suraj Chandrakant Kamble, Dr. Sanmath Shetty, Dr. Sanmath Shetty, Dr. Sonali Khurjekar

Journal of Asia Entrepreneurship and Sustainability · 29.10.2026

DOI: 10.66635/tq2ans20 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Journal of Asia Entrepreneurship and Sustainability, online erschienen am 29.10.2026. Verfasst von Dr. Suraj Chandrakant Kamble und 3 weiteren Autor:innen.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Customer histories are often converted into rolling or adjacent-window features for segmentation and prediction. The same period-level estimate may then appear as an outcome in one panel row and as a predictor in the next, creating apparent dynamics even when no carryover exists. We develop and apply a validation framework combining a no-carryover benchmark, fully disjoint temporal construction, an identical-sample estimator comparison and strict temporal holdout testing. The application uses 102 weeks of grocery transactions and 9,014 household-category relationships observed under both adjacent and disjoint designs. On this identical sample, the adjacent estimator implies a 1.78-percentage-point movement toward full-price purchasing for a 10-point increase in promotional share, whereas the fully disjoint estimator implies a 1.40-point movement in the opposite direction. A 36-cell Monte Carlo grid shows that adjacent estimators can remain positive under zero carryover across panel lengths and event frequencies, while disjoint estimators recover the null; under true persistence, adjacent construction attenuates or reverses the expected sign. Recent promotional share adds only .0063 to holdout AUC. The study shows that temporal feature validation must distinguish predictive usefulness from dynamic interpretation and intervention. The proposed audit supports model governance by identifying when customer-history features are informative for forecasting but unsafe for behavioural or policy claims.

Zitieren

APA

Kamble, D. S. C., Shetty, D. S., Shetty, D. S., & Khurjekar, D. S. (2026). When Temporal Features Reuse Data: Adjacent-Window Bias in Customer Panels. Journal of Asia Entrepreneurship and Sustainability. https://doi.org/10.66635/tq2ans20

Vancouver

Kamble DSC, Shetty DS, Shetty DS, Khurjekar DS. When Temporal Features Reuse Data: Adjacent-Window Bias in Customer Panels. Journal of Asia Entrepreneurship and Sustainability. 2026. doi:10.66635/tq2ans20

BibTeX
@article{Kamble2026,
  author = {Kamble, Dr. Suraj Chandrakant and Shetty, Dr. Sanmath and Shetty, Dr. Sanmath and Khurjekar, Dr. Sonali},
  title = {When Temporal Features Reuse Data: Adjacent-Window Bias in Customer Panels},
  journal = {Journal of Asia Entrepreneurship and Sustainability},
  year = {2026},
  doi = {10.66635/tq2ans20}
}
Kostenlos starten →