Visual Analytics for Machine Learning: Computing and Leveraging Decision Boundary Maps
Francisco Caio Maia Rodrigues · 2020
A machine learning classifier is a program that, given an object, outputs a label indicating its class, among a predefined set of classes. Understanding such classifiers is far from trivial, so designing good ones can be challenging. In this work, we propose a set of visualization techniques that depict how machine learning classifiers effectively partition their data space into decision zones, each one being assigned a different label. We implement and evaluate our visualizations using different techniques, such as dimensionality reduction, inverse dimensionality reduction, dense visualizations, and machine learning classifiers. We complete our work by proposing a visual analytics workflow that can help data scientists to construct and fine-tune their classifiers applied to real-world problems.