Exploring high dimensional data with Butterfly: a novel classification algorithm based on discrete dynamical systems.

Area of Science:

• Computational Biology
• Data Science
• Machine Learning

Background:

• High-dimensional data visualization is challenging without geometric projections.
• Existing methods often rely on principal components or transformed axes.
• Personalized medicine requires methods to explore complex patient data for subclusters.

Purpose of the Study:

• Introduce a novel method for visualizing high-dimensional data using discrete dynamical systems.
• Develop a human-readable 2D representation of subject relationships.
• Detect unrevealed patient subclusters for personalized pathology exploration.

Main Methods:

• Utilizes a discrete dynamical system, related to the chaos game and iterative function systems.
• Employs a memory-type mechanism for data representation.
• Integrates feature selection, dynamical system modeling, and model evaluation for classification.

Main Results:

• The Butterfly algorithm provides a 2D representation of high-dimensional data relationships.
• Successfully applied to public lung cancer dataset and two other datasets.
• Demonstrates capability for both supervised and unsupervised machine learning tasks.

Conclusions:

• Butterfly offers a novel approach to high-dimensional data visualization and classification.
• Facilitates personalized exploration of medical data by detecting patient subclusters.
• The accompanying R package provides a working implementation for practical application.