I am trying to compare different clustering algorithms on a dataset and compare the model performance. Since the dataset is quite big (56 features), I applied PCA to reduce the number of features to just 3 features.
Next, I am trying to evaluate the performance of 3-4 clustering algorithms.
I created silhouette plots to check for the ideal number of clusters. I noticed that although the silhouette plot for BIRCH model with 4 clusters does not seem optimal since there are clusters with negative silhouette scores, the clusters seem good when plotting the principal components.
Here are the screenshots.
Here is the plots between the PCA's with 4 clusters
According to me, I think 4 clusters looks great. But the silhouette plot dhows that this is not the case and that 2 clusters is the ideal number of clusters.
My question is, is silhouette plot suitable for choosing number of clusters for BIRCH models, or any models other than Kmeans clustering?
Searching online, I noticed silhouette plots were only used for kmeans algorithm.