A large US based Pharmaceutical company wants to predict the next 36 months demand for their worldwide products. The objectives were to 1. Improve the demand forecast system’s accuracy and reduce the level of manual intervention, 2. Build an automatic model selection framework which identifies the best model combination (out of 40 time series, machine learning and deep learning models) for each demand category and 2. Setup a ML pipeline on AWS cloud to easily manage a large suit of models and frequently updating them for better results.
An automated Machine Learning pipeline was set-up for seamless production deployment and faster deployment of new changes/models into production. The solution included the following elements:
The ML pipeline for demand prediction system helped to quickly push new approaches and model version into production and complete the project before time.
By setting up end to end pipeline on AWS cloud, overall model forecast time was reduced from 18 Hours to 13 hours (~27% reduction)
Our solution reduced the forecast error (Root Mean Square Error) which helped in improving the production planning and thus reduced the cost