Chapter 14 1) Interpretability and Explainability Machine Learinig

14.1 1.1) Some definitinions

The distinction between interpretability and explainability lies in their focus and depth.

  • Interpretability delves into comprehending the inner mechanisms of the algorithm.

  • Explainability aims to elucidate the decisions is makes.

Model complexity: when dealing with intricate models like Random Forest (with tens of variables and thousands of trees), up to deep neural networks, interpretability becomes challenging due to their complexity and the interplay among their components. In such scenarios, explainability proves to be a more practical approach, as it focuses on clarifying decisions rather than delving into the complexities of the model.

Communication: in terms of audience and purpose, interpretability primarily concerns AI specialists and researchers, whereas explainability targets end users seeking to grasp model decisions. Consequently, explainability necessitates a more straightforward and intuitive communication of information.

14.2 1.2) Aim of the present lab

In this exercise you will work with the outputs of Random Forest resulting from the landslides susceptibility map project (LSM_RF).

  • Firstly, we will explore the relative importance of the predictor variables, and then their relative probability of prediction success.

  • Finally we will apply a local version of Random Forest (named Geographical Random Forest) to analyse the spatial heterogeneity of the local variable importance.

14.3 1.3) Re-load lybraries and workspace

If you have quit the workspace where you have run the RF model for landslide susceptibility map you need to load it again in this new project. “Loading the workspace” refers to the action of restoring the saved state of the R environment. When you save your workspace in R, it typically includes all the objects (such as variables, functions, data frames, etc.) that are currently present in your R session. Loading the workspace means to restore this saved state, bringing back all the previously saved objects into your current R session.

##  [1] "RColorBrewer" "tidyr"        "randomForest" "classInt"     "plotROC"     
##  [6] "ggplot2"      "pROC"         "dplyr"        "readr"        "foreign"     
## [11] "terra"        "distill"      "stats"        "graphics"     "grDevices"   
## [16] "utils"        "datasets"     "methods"      "base"