Clustering forms a major part of showing different relations between data points. Real-time clustering algorithms can visualise relationships between elements in a 3D environment, provide an analysis of data that is separate from the underlying structure and show how the data changes over time.
This paper analyses whether conventional clustering algorithms can be adapted to real-time dynamic data while remaining stable over time. By implementing an agglomerative hierarchical clustering algorithm combined with an exponential decay smoothing function, this paper tested several different distance functions and compared their resulting clusterings. It then derives a stable distance function for clustering sailboat competitors during a regatta and compared different smoothing values to see the impact on the final result.
The paper shows that an adaptively chosen smoothing value provides the best balance between cluster stability and an intuitive visualisation. This paper concludes this solution can be used and adapted to fit a multitude of applications by changing the distance function and the clustering depth.