Optimizing Data Stream Representation: An Extensive Survey on Stream Clustering Algorithms

Question

1. What are the contributions mentioned in the paper "Optimizing data stream representation: an extensive survey on stream clustering algorithms" ?

2. What are the future works in "Optimizing data stream representation: an extensive survey on stream clustering algorithms" ?

3. What is the main challenge for algorithms of this category?

4. What is the common distance measure for binary, ordinal, nominal or text data?

Accepted Answer

This survey explores, summarizes and categorizes a total of 51 stream clustering algorithms and identifies core research threads over the past decades.. This survey is considerably more extensive than comparable studies, more up-to-date and highlights how concepts are interrelated and have been developed over time.. Furthermore, it discusses applications scenarios, available software and how to configure stream clustering algorithms.

Accepted Answer

Future work should systematically benchmark and configure prominent stream clustering algorithms and determine respective strengths and weaknesses, e. g., regarding cluster structure, computational complexity and clustering quality.

Accepted Answer

The main challenge for algorithms of this category is how to construct the grid-cells, i.e., how often cells are partitioned and how to choose the size of cells.

Accepted Answer

for binary, ordinal, nominal or text data, appropriate distance measures such as the Jaccard index, simple matching coefficient or Cosine similarity could be used.

Accepted Answer

For reclustering all selected leaders are used to run DBSCAN.4.4 MedoidsAn alternative to storing Clustering Features is to maintain medoids of clusters, i.e., representatives.

Accepted Answer

Since it is infeasible to store all observations within the clusters, distance-based algorithms usually summarize the observations associated with a cluster.

Accepted Answer

The interplay between the online and offline component is then crucial for decision making since it helps to reveal hidden structures and dependencies within the data streams.

Accepted Answer

Since the weight of cells is calculated relative to the total number of observations, outdated cells can be removed and their statistics returned to the parent cell.

Accepted Answer

In order to keep the number of microclusters constant, outdated clusters are removed based on a threshold on their average time stamp.

Accepted Answer

A popular example is the k-means algorithm which minimizes the within-cluster sum of squares, i.e., the distance from data points to their cluster centroids.

Accepted Answer

The only category which the authors consider out of scope for this paper are stream clustering algorithms for textual data since they usually rely on considerably different approaches.

Accepted Answer

the authors in Nguyen et al. (2015) focus on stream clustering and stream classification and present a total of 17 algorithms.

Accepted Answer

The algorithm normalizes each dimension by regularly sampling the current standard deviation and adjusting the existing clusters accordingly.

Accepted Answer

stream clustering could be applied for market or customer segmentation where customers that have similar preference or behavior are identified from a stream of transactions.

Accepted Answer

While this often helps to achieve better clustering results, it makes algorithms such as LeaDenStream (Amini and Wah 2013), D-Stream with attraction (Tu and Chen 2009) and DBSTREAM less suitable for faster streams.

Optimizing Data Stream Representation: An Extensive Survey on Stream Clustering Algorithms

Chat with Paper

AI Agents for this Paper

Most frequently asked questions

1. What are the contributions mentioned in the paper "Optimizing data stream representation: an extensive survey on stream clustering algorithms" ?

2. What are the future works in "Optimizing data stream representation: an extensive survey on stream clustering algorithms" ?

3. What is the main challenge for algorithms of this category?

4. What is the common distance measure for binary, ordinal, nominal or text data?

5. What is the alternative to storing Clustering Features?

6. What is the common way to summarize a cluster?

7. What is the importance of interplay between the online and offline component?

8. How is the weight of cells calculated relative to the total number of observations?

9. How many microclusters are removed from the database?

10. What is the popular example of k-means?

11. What are the only categories of algorithms which the authors consider out of scope for this paper?

12. How many algorithms are presented in Nguyen et al. (2015)?

13. How does the algorithm normalize the dimensions of a cluster?

14. What is the purpose of stream clustering?

15. What makes the algorithms less suitable for faster streams?

Figures

Citations

When is nearest neighbor meaningful

Automated Algorithm Selection: Survey and Perspectives

Data stream analysis: Foundations, major tasks and tools

A survey on machine learning for recurring concept drifting data streams

A Grid and Density-based Clustering Algorithm for Processing Data Stream

References

A density-based algorithm for discovering clusters a density-based algorithm for discovering clusters in large spatial databases with noise

A density-based algorithm for discovering clusters in large spatial Databases with Noise

Self-organized formation of topologically correct feature maps

k-means++: the advantages of careful seeding

Estimating the number of clusters in a data set via the gap statistic

Related Papers (5)

Density-Based Clustering over an Evolving Data Stream with Noise.

A framework for clustering evolving data streams

A Survey of Stream Clustering Algorithms

Data stream clustering: a review

Data Stream Clustering Techniques, Applications, and Models: Comparative Analysis and Discussion