5
$\begingroup$

I have been doing some research on different type of machine learning (ML) algorithms such as random forest/SVM etc. in order to model and best predict pharmaceutical needs of patients suffering from a particular type of kidney autoimmune disease.

What I was hoping someone could explain to me is what are the differences in predictive ability between Monte Carlo simulations and random forest classifiers? How is their real world application different?

Any comments would be greatly appreciated.

$\endgroup$
  • $\begingroup$ More details are needed here - "Monte Carlo simulation" is something of a catch-all phrase for repeated random draws from a distribution. It is not in an of itself a predictive technique. $\endgroup$ – Fomite Feb 18 '14 at 14:32
2
$\begingroup$

MC is not an inference technique for finding the "best" model, it is a numerical tool to obtain samples from a given model. Sure enough you can also build inference procedures relying on MC (e.g. optimizing a criterion over parameters as a function of the simulated empirical distribution) but that doesn't change the respective scopes and goals. The most common application of MC is probably the calculation of high-dimensional integrals.

$\endgroup$
  • $\begingroup$ You could give us more context on what made you ask such a question, it probably arises from one such application of MC in inference. $\endgroup$ – Quartz Feb 18 '14 at 9:43
  • $\begingroup$ Thanks for both your answer and comment, just digging a little deeper: "a numerical tool to obtain samples from a given model" to increase the amount of data you have for modelling?? If I was going to predict the outcome of any given event, for sake of arguement lets say 'motor car racing' we have all of our required parameters and data representing the last 20 years of motor car races. How would MC be used to enable you to best predict who would win the next motor car race? And would it be more or less accurate than the ML techniques previously mentioned? $\endgroup$ – sunboya Feb 18 '14 at 21:36
  • $\begingroup$ Please read again, the comparison you're asking for is not meaningful. The samples generated via MC are synthetic, artificial, only as good as the model they come from and therefore need not to adhere to the real data, let alone improve over it. It cannot predict anything more than the chosen model. You cannot compare MC to ML techniques: these generate descriptive/predictive models, MC requires a model from the start and does not give you one. $\endgroup$ – Quartz Feb 19 '14 at 12:44
  • 1
    $\begingroup$ You could even generate random outcomes ("MC") from a model estimated by some ML technique (maybe to check afterwards how well the model matches the original data, e.g. comparing realized to synthetic empirical distributione). $\endgroup$ – Quartz Feb 19 '14 at 12:45

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.