Most analytic modelers wait until after they’ve built a model to consider deployment. Doing so practically ensures project failure. Their motivations are typically sincere but misplaced. In many cases, analysts want to first ensure that there is something worth deploying. However, there are very specific design issues that must be resolved before meaningful data exploration, data preparation and modeling can begin. The most obvious of many considerations to address ahead of modeling is whether senior management truly desires a deployed model. Perhaps the perceived purpose of the model is insight and not deployment at all. There is a myth that a model that manages to provide insight will also have the characteristics desirable in a deployed model. It is simply not true. No one benefits from this lack of foresight and communication. This session will convey imperative preparatory considerations to arrive at accountable, deployable and adoptable projects and Keith will share carefully chosen project design case studies and how deployment is a critical design consideration.
- Which modeling approach continues to be the most common and important in machine learning
- The iterative process from exploration to modeling to deployment
- Which team members should be consulted in the earliest stages of predictive analytics project design?
- Misconceptions about predictive analytics, modeling, and deployment
- Costly strategic design errors to avoid in predictive analytics projects
- Common styles of deployment.