Trap Detection in Brownian Particle Trajectories Using Machine Learning Clustering Methods
Lyudmyla Kirichenko, Daryna Khatsko, Оксана Сергеевна Пичугина · 2023
The article is focused on the detection of traps that capture a Brownian particle using machine learning clustering methods. The trajectory of the Brownian particle is simulated using a Brownian motion model with drift, encompassing both free diffusion and trap-bound motion. For temporal data clustering, two density-based clustering methods, DBSCAN and HDBSCAN, are employed. The versatility of these methods allows for cluster identification without prior knowledge of their quantity or shape, making them suitable for trap detection. Through extensive experimentation, the study reveals that the DBSCAN method outperforms HDBSCAN, achieving an average accuracy of 94.0% compared to HDBSCAN’s 85.7%. The effect of model parameters on clustering accuracy was also investigated.