Work

A predictive prescriptive AI framework for sustainable inventory optimization of perishable dairy products: A multi-model forecasting and reinforcement learning approach

Year: 2026

Type: software

Abstract: This repository contains the implementation of a predictive-prescriptive AI framework for sustainable inventory optimization of perishable dairy products, combining multi-model demand forecasting with a PPO-based reinforcement learning agent. Key components in this version: Stochastic delivery lead time environment (L ~ DiscreteUniform{0,1,2} weeks), replacing an earlier zero-lead-time formulation N-BEATS and PatchTST forecasting benchmarks, compared against the existing ensemble forecaster Empirical sample-adequacy analysis (learning curves and training-size ablation) Policy sensitivity analysis under stochastic lead time, verified across 5 independent random seeds Consolidated statistical verification (Wilcoxon signed-rank and Diebold-Mariano tests, with confidence intervals and effect sizes) See README.md for the full repository structure and instructions to reproduce all results.

Source: Zenodo (CERN European Organization for Nuclear Research)

DOI: 10.5281/zenodo.21483881

Authors: Sasi, Archana, Shieh, Chin-Shiuh, Horng, Mong-Fong, Ramu, Arulmurugan, Ravichandran, Sathish Kumar

Institutions: National Kaohsiung University of Science and Technology, Aktobe Regional State University named after K.Zhubanov, Swami Vivekananda Yoga Anusandhana Samsthana

Language: English

Topic: Supply Chain and Inventory Management

Subfield: Management Information Systems

Field: Business, Management and Accounting

Domain: Social Sciences

Open Access status: green

OpenAlex ID: W7170094679

Cites:

Cited by:

Related to: 10