Related Experiment Videos
Comparison of five clustering algorithms to classify phytoplankton from flow cytometry data.
M F Wilkins1, S A Hardy, L Boddy
1Cardiff School of Biosciences, Cardiff, United Kingdom.
Cytometry
|June 29, 2001
Summary
The adaptive distances algorithm, a fuzzy k-means variant, reliably clusters analytical flow cytometric (AFC) data from marine phytoplankton. This method improves upon existing algorithms, though automatic cluster number determination needs further research.
Area of Science:
- Aquatic ecology
- Computational biology
- Data analysis
Background:
- Analytical flow cytometry (AFC) data present clustering challenges for artificial neural networks (ANNs).
- Fuzzy k-means algorithm variants were developed to handle nonspherical clusters using scatter matrices.
- Four variants were proposed, each optimizing different clustering quality measures.
Purpose of the Study:
- To compare the performance of four fuzzy k-means algorithm variants for clustering AFC data.
- To evaluate the effectiveness of these variants against a critical distances algorithm.
- To identify a robust clustering method for ANN training data in aquatic ecology.
Main Methods:
- Four fuzzy k-means algorithm variants were implemented.
- The adaptive distances (Gustafson--Kessel) algorithm was among the variants tested.
- Performance was evaluated using AFC data from cultured marine phytoplankton species.
Main Results:
- The adaptive distances (Gustafson--Kessel) algorithm demonstrated robustness and reliability.
- Other fuzzy k-means variants exhibited various issues during analysis.
- The adaptive distances algorithm outperformed the critical distances algorithm.
Conclusions:
- The adaptive distances algorithm is a superior choice for clustering AFC data compared to tested alternatives.
- Further research is needed to address the automatic determination of the optimal number of clusters.
- This finding aids in the automated extraction of clusters for ANN training data in aquatic ecology.