More and more, data science and decision science practitioners are seeking to combine machine learning forecasts with actionable and optimized decisions. This can include anything from predicting traffic patterns for delivery scheduling to consumer buying behavior for inventory management. But bridging these two disciplines can be challenging.
In the on-demand logistics space, these worlds are colliding more frequently with practitioners generating demand forecasts that feed into shift scheduling models that feed into vehicle routing models. Getting to an 80% good solution for the optimization side is not hard. What is hard is the remaining 20% where people tend to over-optimize their models using fixed inputs (which makes the model more brittle in the face of uncertainty). What if, instead, we could take those 80% solutions and use horizontal compute to scale them up in the face of uncertainty?
In this talk, we will explore what has made blending ML and OR outputs challenging, the roles of deterministic and stochastic optimization in relation to ML, and how scenario testing techniques via horizontal computing provide an expedient and more accessible path for combining these worlds.
This presentation was originally given on June 7, 2024 via the Euro Practitioners' Forum webinar series. It is redistributed with Euro Working Group's permission. Original landing page here: https://www.euro-online.org/websites/or-in-practice/7-june-2024-three-model-problem-combining-machine-learning-ml-and-operations-research-or-through-horizontal-computing/
.png)
