Performance of clustering algorithms on time-dependent spatial patterns

Abstract

A single linkage clustering algorithm adapted for points distributed over time (SLOT) is used to recover clusters from coplanar points associated with time parameters. Cluster number and shape determine separability and hence effectiveness of the algorithm. Performance in simulation experiments also depended on the probability of recording cluster points. SLOT links points observed at different times if they are within some limiting distance, and the distance parameter becomes critical when detectability is low. Performance comparisons are made with other algorithms, and results are presented in the context of a rule-based expert system for solving problems involving cluster analysis of time-dependent spatial patterns.