Functional distributional clustering using spatio-temporal data.

Area of Science:

• Data Science
• Network Analysis
• Spatial Statistics

Background:

• Overcrowded networks present challenges in identifying spatially contiguous clusters with evolving temporal patterns.
• Existing spatially adapted hierarchical clustering algorithms often fail to incorporate temporal dynamics inherent in sensor network data.

Purpose of the Study:

• To introduce a novel non-parametric method, the functional distributional clustering algorithm (FDCA), for identifying clusters in spatial networks.
• To address the limitations of traditional methods by incorporating both spatial and temporal characteristics of network data.

Main Methods:

• The functional distributional clustering algorithm (FDCA) employs an agglomerative hierarchical clustering approach.
• It utilizes a novel distance measure based on cumulative distribution functions of temporal changes across spatial locations.
• The method is fully non-parametric, handling multi-modal distributions of sensor observations.

Main Results:

• FDCA successfully identifies spatially contiguous clusters that incorporate temporal pattern changes.
• The algorithm demonstrated superior performance in identifying true clusters compared to functional-only and distributional-only algorithms.
• FDCA showed comparable performance to a model-based clustering algorithm in empirical and simulated datasets.

Conclusions:

• The functional distributional clustering algorithm (FDCA) is an effective method for analyzing complex spatial networks with temporal dynamics.
• FDCA offers a significant advancement over traditional clustering techniques by integrating spatial and temporal data features.
• The method's applicability is validated through real-world and simulated sensor network data.