Contextual Perishable Retail Demand Forecasting and Inventory Optimization: Benchmarking Hybrid CNN-GRU Against LSTM and ARIMA Models

Chourasia, Kshiteej (2025) Contextual Perishable Retail Demand Forecasting and Inventory Optimization: Benchmarking Hybrid CNN-GRU Against LSTM and ARIMA Models. Masters thesis, Dublin, National College of Ireland.

Preview	PDF (Master of Science) Download (1MB) | Preview
Preview	PDF (Configuration Manual) Download (1MB) | Preview

Abstract

With an estimated value in the excess of 1.3 trillion, the world in the perishable retail market is plagued by enormous wastes, almost 30 percent of the fresh produce and 22 percent of dairy products, largely to do with the unpredictability of demand and rigid inventory policies. Conventional methods of forecasting like ARIMA and exponential smoothing have a tendency to fail to capture the influences of promotions, holidays and other externalities which render them ineffective in a dynamic multi-store retail environment. The current analysis fills these gaps by designing a hybrid deep learning model, where the convolutional architecture and the recurrent architecture (CNN-GRU) take into account external conditions, e.g., weather, local events, and promotions, to enhance prediction of perishable product demand. The balance between the minimization of the perishable wastes and the minimization of the inventory turns by combining these forecasts with a dynamic inventory policy that adjusts the reorder levels to the prediction uncertainty will be achieved through the proposed system.

Through the use of large transaction data of Corporacion Favorita in Ecuador augmented with holiday and weather data in a MongoDB environment, it can be shown that this comprehensive solution can outperform classical ARIMA baselines in terms of accuracy of forecasts resulting in obtaining a mean absolute error of approximately 7.62 units and successfully modeling sales surges around crucial demand times such as holidays. In practice, the benefits amount to increased accuracy of item-level inventory management that has the potential to decrease spoilage and related expenses and enhance stock optimization.

The proposed work will have the benefit of being truly innovative to the academic and practitioner communities and providing evidence of a possible roadmap for integrating real-time forecasts with a responsive inventory policy. In total, the study concludes on the possibility of AI-driven demand prediction solutions to overcome the long-standing issues associated with the management of retail perishable goods and help these businesses remain sustainable and efficient in their operations.

Item Type:	Thesis (Masters)
Supervisors:	Name Email Nagahamulla, Harshani UNSPECIFIED
Uncontrolled Keywords:	Perishable goods retail; Demand forecasting; Deep learning; CNNGRU model; Inventory management; Dynamic reorder policy; External factors; Machine learning in retail; Waste reduction
Subjects:	Q Science > QH Natural history > QH301 Biology > Methods of research. Technique. Experimental biology > Data processing. Bioinformatics > Artificial intelligence Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Artificial intelligence Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Machine learning H Social Sciences > HD Industries. Land use. Labor > Specific Industries > Retail Industry
Divisions:	School of Computing > Master of Science in Data Analytics
Depositing User:	Ciara O'Brien
Date Deposited:	01 Jul 2026 08:34
Last Modified:	01 Jul 2026 08:34
URI:	https://norma.ncirl.ie/id/eprint/9419

Actions (login required)

View Item