Mondrian Embeddings for Visualization of Decision Tree Ensembles.

Area of Science:

• Machine Learning
• Data Visualization
• Bioinformatics

Background:

• Decision trees are crucial for explainable AI in medical diagnosis and classification.
• Traditional decision tree visualizations become ineffective with increasing data complexity and ensemble methods.
• Existing methods struggle to maintain interpretability for high-dimensional and large datasets.

Purpose of the Study:

• To propose a new visualization technique for enhanced interpretability of decision trees and ensembles.
• To address the limitations of current visualization methods for complex datasets.
• To intuitively visualize findings from decision tree models in high-dimensional spaces.

Main Methods:

• Developed a novel visualization technique based on the decision-making process of decision trees.
• The technique visualizes the distinction timing of data pairs to infer data proximity.
• Applied the method to five biology datasets for evaluation.

Main Results:

• The proposed method allows intuitive visualization of low-dimensional data embeddings.
• It effectively visualizes the behavior of the predictor within the same space as the data.
• Demonstrated advantages in understanding decision tree findings for complex biological data.

Conclusions:

• The new visualization technique significantly enhances the interpretability of decision trees and ensembles.
• It provides an intuitive understanding of model behavior and data characteristics for high-dimensional data.
• The method shows promise for applications in bioinformatics and other fields requiring explainable AI.