This document displays a decision-tree fit to the results of the exploratory analysis of the ABM simulation model. This tree was fit using the RPART algorithm. This undertakes a process of recursive partitioning, which splits the data repeatedly on one key parameter at a time. At each split the parameter is chosen which best divides the data into sets of simulations where the failure scenario always did, or always did not, occur. This is known formally as increasing the purity of the terminal nodes of the ‘split’. Because the parameter is chosen relative to which ever subset (i.e. partition) of the data is being split, RPART creates the potential for complex interactions between parameters, such as when a parameter value makes failure more likely in combination with one set of other parameter values, and less likely in combination with some other set of parameter values. The RPART algorithm is used here with default parameters and a complexity penalty of 0.005 (which determines when splitting stops).
Note that the value in the terminal nodes is the probability of the failure scenario occuring