Recurrent Neural Network for Predicting Sequential Supply Chain Delays

This capstone project addressed the challenges faced by GlaxoSmithKline’s (GSK) supply chain in managing sequential delays, essential for ensuring timely healthcare delivery in the pharmaceutical industry. The key objectives included pinpointing planned dates within GSK’s system and developing a robust machine learning model to predict sequential delays accurately. Through an extensive literature review and methodology development, the project focused on utilizing neural network machine learning methods, specifically recurrent neural networks (RNNs) and Long Short-Term Memory (LSTM) networks. Detailed summary statistics showcasing the delay frequencies and locations within GSK operations for the Benlysta brand revealed that approximately 40% of data exhibited delay issues, primarily in primary manufacturing sites. Further examination highlighted specific areas prone to delays, providing GSK with managerial insights for targeted action. The RNN model development involved data acquisition, manipulation, preprocessing, and model construction, followed by hyperparameter tuning to optimize performance, resulting in a reduced mean absolute error (MAE) of 4.89 days. Although challenges in linking manufacturing and quality data limited the initial scope, the project provided valuable insights and laid a solid foundation for future enhancements. Leveraging the findings and insights gained from this capstone project, GSK can enhance operational efficiency, mitigate supply chain risks and deliver medications to patients more effectively.