Session Overview
With constrained supply chains and logistic networks, it has become increasingly difficult to plan operations that are efficient and profitable. In a multimodal transportation process, wide variability in impelling factors leads to long tails in the arrival times’ distributions making it harder to predict. Traditionally prevalent machine learning approaches that provide a single-point prediction of arrival time fall short on accuracy and equally pivotal – stability of estimation.
The solution presented in this talk demonstrates that a more practical estimation of arrival times can be provided by exploiting the leaf nodes on tree-based models combined with a network model. The talk will also cover how to prepare training and test data to minimize data leak. It will demonstrate the details of quantifying the stability metric and optimizing the algorithm for the appropriate ratio of accuracy and window. This approach led to a gain in accuracy and proven efficiency in stable planning.
The Why: A novel machine learning approach to predict arrival times for unusual events, and how to quantify its effectiveness for stable planning.
Key Takeaways:
- How to use real-time feature engineering and handle its challenges.
- Unconventional integration of machine learning and network models
- Designing right metrics and KPIs for arrival time predictions
This session is for… Data Scientists, Data Engineers, Product managers, Business Analysts.
