Feature Selection and Hyperparameter Optimization for Machine Learned Classification of 3D Single-Particle Tracking.

Area of Science:

• Physical Chemistry
• Biophysics
• Materials Science

Background:

• Diffusion in charged and crowded media presents significant challenges for traditional analysis methods.
• Characterizing mixed motion types using traditional techniques like mean square displacement in 3D single-particle tracking (3D SPT) is often difficult.

Purpose of the Study:

• To develop a more accurate method for classifying diffusion behaviors in complex environments.
• To leverage machine learning for analyzing heterogeneous transport phenomena.

Main Methods:

• Employed machine learning, specifically decision tree algorithms.
• Utilized feature selection to identify the most relevant parameters for trajectory analysis.
• Applied the method to analyze diffusion in charged and crowded media.

Main Results:

• Identified six key features crucial for accurate trajectory characterization.
• Demonstrated the effectiveness of machine learning in distinguishing between different diffusion types.
• Successfully classified complex diffusion patterns that are challenging for traditional methods.

Conclusions:

• Machine learning, particularly decision trees, offers a powerful tool for understanding heterogeneous transport.
• This approach advances the study of diffusion in challenging biological and materials science contexts.
• The identified features provide insights into the dynamics of charged and crowded systems.