Simple and efficient for large datasets.

Works well with spherical clusters.

Computationally fast compared to other methods.

Requires specifying the number of clusters (k).

Struggles with non-spherical clusters and varying cluster sizes.

Sensitive to initial centroid placement and outliers.

No need to predefine the number of clusters.

Produces a dendrogram, allowing flexibility in choosing the number of clusters.

Computationally expensive (O(n²) or O(n³)), making it unsuitable for large datasets.

Sensitive to noise and outliers.

Cannot adjust once a merge or split is done.

Can find clusters of arbitrary shapes.

Does not require specifying the number of clusters.

Handles noise and outliers well.

Struggles with varying density clusters.

Performance decreases with high-dimensional data.

Requires careful tuning of parameters (ε, MinPts).

No need to predefine the number of clusters.

Can detect arbitrarily shaped clusters.

Computationally expensive for large datasets.

Bandwidth selection is crucial and non-trivial.

Can model complex cluster shapes using probability distributions.

Provides a probabilistic measure of cluster membership.

Requires specifying the number of clusters.

Computationally expensive compared to K-Means.

Can converge to a local optimum.

Works well for non-convex clusters and graph-based data.

Can capture global structure.

Computationally intensive (depends on eigenvalue decomposition).

Not scalable for large datasets.