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

Understanding diffusion in charged and crowded media is crucial for solving a wide range of biological and materials challenges. Classifying diffusion by traditional methods such as mean square displacement in three-dimensional single-particle tracking (3D SPT) is difficult, especially when there are mixed motion types. To address this, we employed machine learning (ML), specifically decision tree algorithms with feature selection, to identify the six most relevant features for accurate characterization of trajectories. This work demonstrates the value of ML in advancing our understanding of heterogeneous transport that occurs in charged and crowded environments, with a broad range of applications.