Machine Learning and Stochastic Simulation for Inventory Management

This thesis explores the use of advanced data-driven techniques for dimensioning safety stock and optimizing inventory in a supply chain. The thesis is based on data and insights for raw material inventory at Amgen, a biotech company. Resilient inventory management is important in the biopharma and biotech sector as the repercussions of a drug shortage are dire. However, the complexity of biomanufacturing processes creates significant variability and uncertainty around lead times and demand. Amgen currently holds high raw material inventories across thousands of materials to mitigate risks of stockouts that could delay production. However, the policies of holding high raw material inventories in Amgen have resulted in increased holding costs and also tied up working capital. To address this challenge and find a sustainable method for managing raw materials in the company and by extension, other stages of production, a novel methodology is developed. Machine learning models such as CatBoost, Extreme Gradient Boosting (XGBoost) and Random Forest are proposed to forecast lead times and demand. The models are trained on datasets of 10,000+ materials, incorporating unique patterns based on factors like suppliers’ historical delivery performance, historical demand pattern and material characteristics. A segmentation framework is also developed to properly allocate service levels based on risk tolerance for different category of materials. Stochastic simulation then applies the learned predictive distributions to quantify optimal safety stock levels under uncertainties. This considers desired service levels, holding costs, risk tolerance, cost-risk tradeoffs and potential disruptions in what-if scenario cases to support resilience. The methodology is validated on sample materials with both short and long lead times. Results indicate potential inventory reductions of over 25% while still preventing stockouts, enabling multimillion dollar savings in procurement and holding costs. A phased implementation plan is also proposed in order to ensure smooth transition using this new data-driven approach in the organisation, taking into consideration change management. This solution fuses predictive analytics with simulation and optimization to transform safety stock calculation from a cost burden to a competitive advantage. The dynamic data-driven framework significantly enhances supply chain resilience and efficiency in the vitally important biopharmaceutical industry, where patient outcomes are at stake. The methodologies developed could be applied across various production stages and tailored to other sectors.

Name thesis:degree_name

Master

Department dc:contributor.department

Massachusetts Institute of Technology. Department of Electrical Engineering and Computer Science

Grantor dc:publisher

Massachusetts Institute of Technology

Year dc:date.issued

2024

Author dc:creator

• Olaleye, Ololade

Advisors dc:contributor.advisor

• Boning, Duane S.
• Golrezaei, Negin

Statement dc:rights

• In Copyright - Educational Use Permitted
• Copyright retained by author(s)

Licence dc:rights.uri

• https://rightsstatements.org/page/InC-EDU/1.0/

Handle dc:identifier.uri

https://hdl.handle.net/1721.1/155988

OAI identifier oai:identifier

oai:dspace.mit.edu:1721.1/155988