Predictive ML: How Unilever is Improving its Forecasting
OVERVIEW
EXPERIENCE | In Person |
---|---|
TYPE | Breakout |
TRACK | Data Science and Machine Learning |
INDUSTRY | Manufacturing |
TECHNOLOGIES | AI/Machine Learning, Delta Lake, MLFlow |
SKILL LEVEL | Intermediate |
DURATION | 40 min |
DOWNLOAD SESSION SLIDES |
Forecasting has always been a hot topic for Unilever, just as for any other company out there. We are showing how we can improve the existing forecasting processes by employing ML predictive models to predict various key business metrics at different levels of granularity with the objective of providing robust automated forecasting that a) informs, challenges and ultimately improves manual forecasts where these exist and b) fills the gaps where manual forecasts don't exist. The engine behind this is a reusable predictive framework we have created and developed in our Databricks Platform that can be reapplied to different time series forecasting scenarios to gain development efficiencies and easier maintenance. The framework consists of libraries that look after data preparation, feature engineering and selection, ML and statistical model training and deployment, prediction generation, explainability, performance monitoring, and evaluation.
SESSION SPEAKERS
Bill Tsiligkiridis
/Europe Data Science Lead
Unilever