totalvis: A Principal Components Approach to Visualizing Total Effects in Black Box Models

While a wide variety of machine-learning techniques have been productively applied to diverse prediction tasks, characterizing the nature patterns and rules learned by these has remained difficult problem. Often called ‘black-box’ models for this reason, visualization become prominent area research in understanding their behavior. One powerful tool summarizing complex models, partial dependence plots (PDPs), offers low-dimensional graphical interpretation evaluating effect modifying individual predictors on fitted/predicted values. Nevertheless, high-dimensional settings, PDPs may not capture more associations between groups related variables outcome interest. We propose an extension based idea grouping covariates, interpreting total effects groups. The method utilizes principal components analysis explore structure several assessing approximation function. In conjunction with our diagnostic plot, totalvis gives insight into group covariates can be used situations where appropriate. These tools provide useful approach pattern exploration, as well natural mechanism reason about potential causal embedded black-box models.