0

"I currently have 10 users, each with 5 features. The first feature is 1-dimensional, the second feature is 2-dimensional with a shape of [8,2], the third feature is 1-dimensional, the fourth feature is 2-dimensional with a shape of [6,2], and the fifth feature is 2-dimensional with a shape of [9,2]. In this scenario, can I cluster these users based on these features? Which clustering method should I use? Can anyone provide me with some advice?"

i wonder if there is any other method other than using method like PCA to lower the dimension of features to the same shape.After dimension reduction, what kind of clustering algorithm can be used ?

Run Zhang
  • 3
  • 2

2 Answers2

1

I think you can cluster the users based on these features. However, since the dimensions are different for each entity, it would be necessary address this issue before applying clustering algorithms.

One option is dimensionality reduction techniques such as PCA that can be applied to the second, fourth, and fifth features to lower their dimensions to match the first and third features.

Besides PCA, there are other dimensionality reduction methods that you could consider, such as t-SNE, LLE or autoencoders.

Soheil Babadi
  • 562
  • 2
  • 4
  • 15
1

For alternatives to PCA, I think autoencoders could be helpful, as the other answer mentioned. Autoencoders are neural networks that are used to reduce the dimensionality of data. They will first learn an encoded (compressed) representation of the input data, which captures the main characteristics of the data. Then, the decoder part of the autoencoder then reconstructs the original data from the compressed representation.

Ilex
  • 31
  • 3
  • so in this case we are just using encoder part I guess? Since I just want to reduce the dimension of the features? – Run Zhang Jul 12 '23 at 04:37