I have a 15-dimensional data set, how many types of attacks should be used to cluster the data set.
1. Now I have aggregated my data set into 5 clusters (5 attacks). Does anyone know how I can point out which cluster is which attack? (How to label the cluster is not just “cluster 1, cluster 2…”)
2. In supervised classification, we have a training data set and a test data set, and use the classifier built from the training data set For testing. My question is, can the same method be used for clustering. Just like using a clustering algorithm to build a model, and then automatically classify the new instance into a specific cluster? Can this be achieved?
The artificial name is not in the data!
For some clustering algorithms, you can automatically assign new instances, but usually you cannot (don’t know the model used by the clustering). In the worst case, a new observation can even The two clusters are merged into one. What are you going to do?
If you want to classify, use classification instead of clustering.
Clustering has a very different mindset. If you look at it from a classification perspective, you won’t really understand It. You use clustering to find what is unknown in the data, and use classification to summarize what is known in the new data.
If necessary, you can also train a classifier on the cluster. But don’t blindly Do this. First make sure that the clusters are actually useful. Compared to good clusters, it is much easier to come up with completely meaningless clustering results. Training a classifier on worthless clusters will not produce meaningful output.
I am very confused about the following two questions:
I have a 15-dimensional data set, how many types of attacks should be used in the clustering data set. < p>
1. Now I have aggregated my data set into 5 clusters (5 attacks). Does anyone know how I can point out which cluster is which attack? (How to label the cluster is not just “cluster 1, cluster 2…”)
2. In supervised classification, we have a training data set and a test data set, and use the classifier built from the training data set For testing. My question is, can the same method be used for clustering. Just like using a clustering algorithm to build a model, and then automatically classify the new instance into a specific cluster? Can this be achieved?
How to identify naming attacks through unsupervised methods?
The artificial name is not in the data!
For some clustering algorithms, you can automatically assign new instances, but usually you can’t (don’t know the model used by clustering). In the worst case, a new observation can even for example change The two clusters are merged into one. What are you going to do?
If you want to classify, use classification instead of clustering.
Clustering has a very different mindset. If you look at it from a classification perspective, you won’t really understand It. You use clustering to find what is unknown in the data, and use classification to summarize what is known in the new data.
If necessary, you can also train a classifier on the cluster. But don’t blindly Do this. First make sure that the clusters are actually useful. Compared to good clusters, it is much easier to come up with completely meaningless clustering results. Training a classifier on worthless clusters will not produce meaningful output.