Random forests are a popular method for statistical applications with an objective of prediction. While an individual tree within a random forest provides a clear decision path for how a prediction is made, the ensemble of trees from the forest creates a complex decision process that is difficult to interpret. One approach used to gain insight into this decision process is visualization of the model. Various approaches have been taken to visualize random forests including trace plots developed by Simon Urbanek (https://link.
Read more →
Machine learning models are excellent predictors, but it is impractical to interpret many of these models. Despite this impracticality, it is important to be able to explain predictions to assess and validate models. As a result, a field of research has recently developed in the explainability of machine learning models. In this talk, I will provide an overview of explainable machine learning with a focus on visualization methods. I will discuss philosophies of “explainability”, model agnostic and model specific visualization methods, and code for creating some of the visualizations in R.
Read more →
