A novel concurrent approach for multiclass scenario discovery using Multivariate Regression Trees

Abstract

To support equitable planning, model-based analyses can be used to explore inequality patterns arising from different scenarios. Scenario discovery is increasingly used to extract insights from ensembles of simulation. Here, we apply two scenario discovery approaches for unraveling inequality patterns and their drivers, with an application to spatial inequality of farms profitability in the Vietnam Mekong Delta. First, we follow an established sequential approach where we begin with clustering the inequality patterns from the simulation results and next identify model input subspaces that best explain each cluster. Second, we propose a novel concurrent approach using Multivariate Regression Trees to simultaneously classify inequality patterns and identify their corresponding input subspaces. Both approaches have comparable output space separability performance. The concurrent approach yields significantly better input space separability, but this comes at the expense of having a larger number of subspaces, requiring analysts to make extra effort to distill policy-relevant insights.