I have a very large set of animal migration data, consisting of many series of vectors - each series is basically a path of a single animal. The dataset I'm using consists of 244 of these series.

Blue vectors are between pink points, pink points are snapshots of an individual animal's location

I want to train a predictive model so that when it is given a collection of these series and a map of environmental variables, such as a map of ocean surface temperature, it can output a new collection of these series.

My question is which machine learning algorithm would be optimal for this kind of prediction - I was looking into multinomial logistic regression, but that is for categorical data. I want the algorithm to develop a model that thinks at a point with an ocean surface temperature of 20˚C, an animal will travel along a vector with a magnitude of 35km and an angle of 34˚'.

I was also looking into using the random forest algorithm, but that seems to be more for classification than forecasting. Does the algorithm I'm thinking of exist? Thanks so much.

EDIT: My particular question is not how to fit this data with a particular ML method, but is there an ML algorithm that can predict a new vector (in polar form) given an old vector and an environmental variable at the tail of that vector.

  • 1
    $\begingroup$ Possible duplicate of Using the random forest algorithm to predict vectors $\endgroup$ – Tom van der Zanden Oct 8 '15 at 13:34
  • $\begingroup$ It looks like you are basically re-posting your prior question, but with a little bit of a different emphasis. Please don't do that -- we don't want duplicates, and this is not an appropriate way to try to gather more attention to your question. Instead, you should edit your old question by clicking the "edit" link underneath it, to make it reflect what you want to ask. $\endgroup$ – D.W. Oct 8 '15 at 20:19
  • 1
    $\begingroup$ 1. I think one thing that is missing from this question is how you will evaluate the output of your model. What criteria/objective function/metrics will you use to evaluate the output of a proposed model? 2. It sounds like you want more of a generative model rather than a discriminative model. A random forest is a discriminative model rather than a generative model, so not the obvious approach. So, my advice is to spend some time with a ML textbook looking at different kinds of generative models. $\endgroup$ – D.W. Oct 8 '15 at 20:20
  • $\begingroup$ @D.W. What do you mean by 'evaluate the output' of my model? - I plan to test the models accuracy by feeding in temperature data that already has a known migration pattern, then comparing the generated outcomes with those current day migration patterns. Also, thanks. $\endgroup$ – jeshaitan Oct 8 '15 at 20:27
  • 1
    $\begingroup$ We can't tell you what similarity metric makes sense for your application; only you can tell that. Your first step, before you should start thinking about machine learning algorithms, is to specify the problem: to identify a metric to quantify the usefulness of a prediction (e.g., its accuracy). I don't recommend searching around for random notions of graph similarity. Instead, I'd recommend you start with how you are going to use the predictions, and use that to try to come up with a way to quantify the error/inaccuracy in a candidate prediction. $\endgroup$ – D.W. Oct 11 '15 at 10:09

Your Answer

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

Browse other questions tagged or ask your own question.