Interactive slice visualization for exploring machine learning models

TL;DR

This paper introduces an interactive visualization technique using slices of predictor space to improve interpretability of complex machine learning models, implemented in the R package condvis2.

Contribution

It presents a novel interactive visualization method for exploring and understanding machine learning models through slice-based visualization.

Findings

Enables interrogation and validation of models

Supports comparison of different models

Implemented in the R package condvis2

Abstract

Machine learning models fit complex algorithms to arbitrarily large datasets. These algorithms are well-known to be high on performance and low on interpretability. We use interactive visualization of slices of predictor space to address the interpretability deficit; in effect opening up the black-box of machine learning algorithms, for the purpose of interrogating, explaining, validating and comparing model fits. Slices are specified directly through interaction, or using various touring algorithms designed to visit high-occupancy sections or regions where the model fits have interesting properties. The methods presented here are implemented in the R package \pkg{condvis2}.