When studying for a doctoral degree (PhD), candidates submit a thesis that provides a critical review of the current state of knowledge of the thesis subject as well as the student’s own contributions to the subject. The distinguishing criterion of doctoral graduate research is a significant and original contribution to knowledge.
Once accepted, the candidate presents the thesis orally. This oral exam is open to the public.
Forecasting demand effectively and managing inventories efficiently are critical components of modern supply chain management. By understanding the full scope of demand possibilities, businesses gain the ability to fine-tune inventory levels, navigate situations involving stockouts and overstock, and move toward a more resilient and precise supply chain. The focus of this thesis is on the retail industries, exploring which strategies and methodologies can enhance these critical aspects.
We start with examining the impact of customer segmentation on forecasting precision by introducing a novel cluster-based demand forecasting framework that harnesses ensemble learning techniques. Our results showcase the effectiveness of the clustered-ensembled approach with minimal forecast errors. However, data availability and segmentation limitations point to potential areas for future research.
The significance of demand accuracy becomes most apparent when we consider its impact on safety stock. Per the second objective of the thesis, we delve into the application of multivariate time series, forecasting for optimal safety stock and inventory management. We employ deep learning models in an ensemble forecasting approach, followed by a cost optimization framework. This strategy outperforms individual models, demonstrating enhanced forecasting accuracy and stability across diverse product domains. Calculating safety stock based on proposed demand prediction framework leads to optimized safety stock levels. This not only prevents costly stockouts but also minimizes surplus inventory, resulting in reduced overall holding costs and improved inventory efficiency.
Although the first two objectives provided optimized results, relying on point predictions to calculate safety stock is not ideal because safety stock calculations require the average demand across time intervals. Unlike traditional point forecasting, distribution forecasting aims to cover the entire range of potential demand outcomes, essentially creating a comprehensive map of possibilities. The third objective of this thesis introduces recurrent mixture density networks (RMDNs) for refined distribution demand forecasting and safety stock estimation. These innovative models consistently outperform traditional LSTM models, offering more precise stockout and overstock predictions. This approach not only reduces inventory costs but also enhances supply chain efficiency.
In summary, this thesis provides valuable insights and methodologies for businesses seeking to improve their demand forecasting accuracy and optimize inventory management practices using the lenses of customer segmentation, ensemble deep learning, and distribution forecasting. By adopting these innovative approaches, organizations can enhance decision-making processes, reduce operational costs, and ultimately thrive in the ever-evolving landscape of supply chain operations.