Chapter 17 6) Conclusions and further analyses

In the present exercise GRF has been used as a purely exploratory tool to estimate the spatial variation of the relationship between landslides in canton Vaud (Switzerland) and the influencing factors. It allowed to elaborate maps delineating the local average importance of the most highly correlated features and to visualise the local fitting performance (R2 local value) into a map.

To be sure that everything is perfectly clear for you, we propose you to answer the following questions and to discuss your answers with the other participants to the course or directly with the teacher.

1. Among the following angorithm evaluate them in therms of their interpretability and explainability: Support Vector Machines , linear regression, Deep Learning Models, Decision Trees, K-Nearest Neighbors, Neural Networks, Random Forests, logistic regression.

2. Which are the three most important variables of your model (based on the MDA)?

3. What is the slope value (or range of values) that gives the highest probability of landslides occurrence? And for the geology, which are the most important classes?

4. Evaluate the spatial variation of the relationship between landslides and slope / distance to roads in your study area by visually inspecting the local average importance of these features.

5. You can replicate this code (some chiuncks of it) to evaluate the local average importance of the third most important variable, as well as to map the local mean squared error.