I'm struggling to model a machine learning problem and created a stylized version of the problem as a question:

Predicting Future Zombies

It was closed as off-topic. Any suggestions on how to improve the question?

Based on https://stats.stackexchange.com/help/on-topic, the question seems to fall within the machine learning and data-driven computing categories that are considered on-topic.

  • 4
    $\begingroup$ "Good way of modeling this problem" is already so broad and vague as to render the question unlikely to be a good fit for this site. It can be difficult to formulate a clear, precise question when you're in a confusing or inherently ambiguous situation, but it's still worth the effort. Sycorax provides fundamental advice: ask the question you actually confront. Efforts to abstract questions mathematically or to describe the situation through analogies usually backfire because often essential information is obscured, lost, or even corrupted. $\endgroup$
    – whuber Mod
    May 16, 2023 at 17:05

1 Answer 1


I didn't vote to close, but a read of the question doesn't make it clear what the problem is and what, specifically, you need help with. Zombies don't exist in reality, so explaining what problem you're trying to solve in explicit detail is an important part to getting useful answers.

The best course of action is to explain what real-world problem you're facing & where you need help.

Be specific!

I've broken down the question into its (few) sentences with commentary. But the common theme is that it's not clear (1) what the statistical problem is and (2) what part you need help with.

I'm trying to predict the number of zombies we're going to need to kill each day.

Well, there's a wide range of options, including killing 0 zombies or all the zombies. Is there a utility function associated with killing a zombie? Are there constraints? Sending out a party to kill a zombie creates risk for the party & they themselves could be turned, so maybe you need to allocate these resources carefully.

I have historical data on the date each human was bit, the date they turned (if they already turned), and which state they were in so we can prepare the appropriate National Guard unit.

Ok, so there's some data about each zombie, and one can surmise that National Guard units might be doing some (or all?) zombie-killing, but we need some more explicit information on how these things are related.

Given data like this that will contain bites up to the previous day, I'd like to predict the number that will turn in each state on each of the upcoming 60 days.

Is there a 100% conversion rate from bites to zombies? Is the time from bite-to-zombie conversion fixed? Is it a random value arising from a known distribution? Or is it something to be estimated?

Likewise, is your knowledge of bites god-like — total and comprehensive? Or do you only have data on some bites and need to estimate how many people are bitten but not recorded in your data?

I'm struggling to model a machine learning problem and created a stylized version of the problem as a question:

This Meta post is the first time that machine learning is mentioned as being related to your question. What aspect of this is related to machine learning, as opposed to, say, analysis?

Based on context, this is what I think you're asking.

A person got infected with zombie virus on some date. Everyone who is infected eventually turns to a zombie, but we don't know how long after infection a person will turn to a zombie. Given a list infection dates and durations (& some of the durations may be missing because the person with an infection is not yet a zombie), can we estimate the time-to-zombie-ism?

This problem seems to be about times: you have a censored random variable but you know that eventually everyone who is bitten turns to become a zombie, and you want to estimate the duration between the bite and the zombie-ism.

But this is just a guess! My guess could be wrong! You have to explain the problem that you're trying to solve to receive useful answers.


You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .