Forecasting with Machine Learning
Time-series forecasting is one of the longest-standing applications of machine learning, and is one of the most prevalent techniques used across all of industry (if not the most prevalent). And yet, during the recent ML boom, forecasting has been somewhat left behind.
The goal of this course is to marry the latest-and-greatest of the field of ML with the existing, classical statistical techniques. In particular, the focus of this course is the practical applications of these techniques, and how they supercharge applications such as causal analysis, demand intelligence, and labor planning.
Course taught by expert instructors
Mark Tenenholtz
VP of Data Science, Stealth
Mark got his start in forecasting while working for Kroger/84.51 in targeted marketing before moving to their central forecasting team. There, he built forecasting models at scale for the largest grocer in the U.S. While there, he became a Kaggle Competitions Master, including a solo gold medal in a forecasting competition. Now, he is the VP of Data Science for a stealth forecasting startup.
The course
Learn and apply skills with real-world projects.
Data scientists who have a background in forecasting, but want to catch up with the state-of-the-art.
New and experienced data scientists/ML practitioners who want to get their start in forecasting.
Anyone across the data stack (data scientists + engineers) who want to better understand the forecasting models needed to power downstream applications.
Intermediate knowledge of Python, including Pandas and NumPy.
Basic fundamentals of machine learning.
Try these prep courses first
- Learn
- End-to-end overview of forecasting problems
- Making the most of Pandas for time-series data
- Finding signal in time-series data
- Overview of modeling approaches, from univariate to global
Build a category/store/state-level forecasting model for retail store sales- Explore the data at the item level
- Discover the pros and cons of different types of models at different levels
- Decompose retail sales into components
- Learn
- Quantifying model performance
- Setting up reliable backtesting frameworks
- Model interpretation
- Applying ML models to forecasting problems
Build a cross-validation setup and train ML models- Compare different metrics (what do they catch and not catch?)
- Build features, analyze your models, and repeat
- Learn
- Hierarchical forecasting
- Ensembling models
Combine what you learned in weeks 1+2 to make even better models- Create optimized ensembles for your high-level and low-level models
- Decompose your high-level models into low-level predictions, aggregate your low-level models to high-level predictions, and reconcile the two for great performance
- Compare the robustness of ensembles to individual models over multiple time periods
- Learn
- Use-cases for time-series models, including causal analysis, labor planning, and understanding demand drivers
- Causal analysis: understand the the effect of business decisions
- Labor planning: using forecasting to properly staff retail stores
- Demand intelligence: interpret your models to understand what drives demand, why some locations succeed (and others don’t), and even help choose new locations
- Building a pipeline to retrain your models
Project details coming soon...
Real-world projects
Work on projects that bring your learning to life.
Made to be directly applicable in your work.
Live access to experts
Sessions and Q&As with our expert instructors, along with real-world projects.
Network & community
Core reviews a study groups. Share experiences and learn alongside a global network of professionals.
Support & accountability
We have a system in place to make sure you complete the course, and to help nudge you along the way.