2
$\begingroup$

I am new to Machine Learning. For Iris data set problem we can solve the problem with Multinomial logistic regression and as well as neural network. Which would give better performance regarding with cost and error?

$\endgroup$
1
  • 5
    $\begingroup$ As stated, this question is too broad to be answerable. Can you clarify "cost and error" to make your Q more specific? $\endgroup$ Nov 7, 2013 at 19:42

1 Answer 1

2
$\begingroup$

As far as I know, the Iris data set should be (almost) linearly separable.

Multinomial logistic regression (MLR) is a linear classifier. Neural networks (NN) are nonlinear classifiers.

The problem with NNs is that they could overfit your training data and might not generalize as good as MLR. You can avoid that by adding a regularization term to the cost function (error function) of the NN, so that your error function will consist of a term that penalizes errors on the training set (e.g. cross entropy, sum of squared errors $\sum_n ||y^{(n)}-t^{(n)}||^2_2$) and a term that penalizes the model complexity (e.g. norm of the weight vector $\gamma||w||^2_2$, ...).

Adding a penalty for large weights to your error function is like using a prior $p(w)$ for your hypothesis. Which means some $w$ become more likely than others.

$\endgroup$

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.