An Application of LIME to a Random Forest Model
Talk for ISU graphics group
Random forests are known for their accurate predictive abilities, but they are a part of the family of machine learning models that lack interpretability. A technique called LIME was developed to provide local interpretations for black-box predictive models. In this talk, I will explain the LIME procedure and show an application of LIME to predictions from a random forest model fit to a bullet matching dataset. I will present a Shiny app I created to view the LIME explanations. Additionally, I will discuss the issues that I have encountered while working with LIME, some of the attempts at a solution, and future directions for my research.
