FORESTER – Interactive visualization of tree-based machine learning

Visualizations are crucial for machine learning as they allow practitioners to understand, analyze, and communicate their models. They help interpret complex models by providing a graphical representation of both data and model performance. Visualizations can be used to understand the underlying patterns and trends in the data, identify biases and errors, and diagnose problems with the model. They also help in communicating the results of the model to a non-technical audience by providing an intuitive and interactive way to present the findings.

Tree-based machine learning methods, such as Classification and Regression Trees or Random Forest, are well-established and widely used in the Earth Sciences. However, visualization tools provided by common machine-learning environments in Python, R, or Matlab often provide graphical representations that could be more visually appealing or helpful in conveying a clear message.

Here we present FORESTER, a web-based and open-source software that produces visually appealing tree-based visualizations. Forester produces publication-ready plots that are, at the same time, interactive figures that can guide the user in interpreting the model. Visualizations can be streamlined to the user's requirements and offer a wide variety of insightful techniques. This makes Forester a promising alternative to currently used environments. Forester is open to collaborations, so we hope it will be extended within the Earth Science community and beyond, proving useful in other machine-learning-related fields.