I ran into a 2019-Entrance Exam question as follows:

Which of the following algorithm has the higher bias?

  1. GMM
  2. GMM (identity covariance matrix)
  3. spectral clustering
  4. k-means

The answer mentioned is (4), but some search on google showed me maybe (1) and (2) is equal to (4). Why would k-means be the algorithm with the highest bias? (Can you please also provide references to valid material to study more?)

  • $\begingroup$ I would say because it is the model with the more reduced flexibility in terms of shape of clusters than can fit. K-means is a hard clustering techniques favouring spherical shapes. I do not know if there is really a quantitative / not qualitative to this question, but this is a better place to make this question I think: stats.stackexchange.com $\endgroup$
    – Thomas
    Jan 19, 2021 at 11:43


You must log in to answer this question.