0
$\begingroup$

I am doing clustering in Weka for a school project. I am trying to compare two Weka outputs with log-likelihood:

Number of clusters selected by cross validation: 6

Number of iterations performed: 13

Clustered Instances

0 54 ( 26%) 1 39 ( 19%) 2 43 ( 21%) 3 20 ( 10%) 4 10 ( 5%) 5 39 ( 19%)

Log likelihood: -52.97762

I am trying to understand if a Log likelihood of-52.97762 is better than -15.671233. Or what's the best Expectation Maximization model if the log-likelihoods are -53, -16 and 10

$\endgroup$
  • 3
    $\begingroup$ Rule #0 of statistical software is to apply it to datasets with known properties before you apply it to the data that concern you. That will help assure you that (a) the software works correctly and (b) you understand its output. In light of rule #0, consider applying Weka to a couple of related small datasets where you know, intuitively, which one is more clustered than the other. (Alternatively, apply it to published datasets with published analyses.) $\endgroup$ – whuber Jan 12 '18 at 15:31
0
$\begingroup$

-15.671233 is better than -52.97762. The log-likelihood in this case, is the probability of your data given the estimated model parameters. The higher the probability, the better the fit. The reason they're negative (rather than between 0 and 1) is because of the logarithm.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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