Table 2 The properties and applicability of the five clustering approaches.
Clustering methods | Applicable scope | Advantage | Disadvantage | Computational complexity | Representative algorithm |
|---|---|---|---|---|---|
Partitioning method | Medium size set | • Easy and efficient for huge data sets • With minimal time and space complexity | • The outcome is simple to be locally optimum • The K value must be predetermined. • Sensitive to the number of K points chosen | O(n) | FCM, K-means, CLARANS |
Hierarchical method | Small quantity set | • Excellent interpretability. • Generates high-quality clusters • May be utilized for non-spherical families | • The intricacy of time is tremendous | O(n2 log n) | CURE, BIRCH, CHAMELEON |
Density-based method | Any family structure | • Insensitive to noise • Can detect groups of any form | • The clustering outcomes are tightly tied to the parameters • Sparser clusters or classes that are closer together perform worse. | O(n log n) | DBSCAN, OPTICS, DENCLUE |
Gird-based method | The inherent data density is low | • High velocity | • Parameter dependent • Unable to handle data with uneven distribution • The algorithm’s efficiency comes at the expense of the accuracy of the clustering findings | O(n) | STING, CLIQUE, WaveCluster |
Model-based method | High-dimensional data | • The outcomes are more visible | • The computational complexity is high • Ineffective execution • When the amount of data is little, it does not operate properly | O(n2) | GMM, SOM, COBWEB |
