I m studying visual analytics and i have a theoretical question about this topic.

My professor introduced this schema in him slide

enter image description here

For connect data to visualisation. Some topic is very easy to understand and regards the rappresentation of data, plot and other tool, other topic instead regards the "data" in particular the data reduction using algorithm like PCA.

I m not very clear why using PCA over the data for reduction some data that we want rappresent. Ad example, if i have dataset with all movie(it s just an example) with some attribute ad example

id,title,year,review, userReview ecc...

Where is the correlation of this data?Why i need apply a PCA for reduce big database and lose some of this components that I can visualize in some plot or schema that for me is important For example could be interesting see all best movie that have high average review, using PCA we just cluster user in different movie (i saw example on Iris database here: https://medium.com/@kyasar.mail/pca-principal-component-analysis-729068e28ec8 )

Anyone can clarify me?


1 Answer 1


You have to see things for what they are. First, you have to figure out what PCA does. It's not clustering, but it can be used to realize that there could be clusters in the data. PCA is not always the answer. It depends on the data. There are other dimensionality reduction techniques such as autoencoders, non-negative matrix factorization and so on...

For the movie example, it does not make sense to use PCA. The iris database contains another type of data, where it does make sense. For the movie example, you might want to have a look at singular value decomposition and collaborative filtering.

One of the reasons for doing PCA could be for clustering. If you have many dimensions, and the intrinsic dimensionality is low you will have the curse of dimensionality. Reducing the dimensionality by PCA and doing clustering on the embeddings solves, in this case, the curse of dimensionality.

  • $\begingroup$ I already do SVD for big data analysis in other course and i had challange with this stuff , all reduction (PCA,SVD, MDS, TSNE , ecc) but in this course is mandatory take a dataset and use a PCA and data visual (in D3). The professor link reduction with visualizzation , but i dont understand how $\endgroup$
    – theantomc
    Feb 24, 2019 at 8:59

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.