4

Consider the problem where I need to predict when a particular event is going to occur based on the past data (data available with time stamp) available.

For example

Assume a particular machine is running and we have the data collected from that particular machine and other similar kind of machine on various time stamps. The data include values collected from a machine which has failed on some functionality and also which is working properly.

What I need to do is to predict WHEN the particular machine(or some functionality) is going to fail.

From the problem What i understood is it is a time series analysis problem. I have heard about survival analysis where outcome is time. But I am not sure which other models I can try.

Could someone please explain which all models can be used for the particular problem (Assume i will be comparing the various quality matrices for models). I there are any approaches other than time series, please advise (My assumption is survival analysis is part of Time series analysis)

Technology to be used will be R or Python (scikit-learn)

Thanks in advance

quemeful
  • 9,542
  • 4
  • 60
  • 69
B K
  • 723
  • 8
  • 17
  • similar question and possible answer here http://stackoverflow.com/questions/36567563/how-to-use-machine-learning-to-calculate-a-graph-of-states-from-a-sequence-of-da/36589044#36589044 – JARS May 11 '16 at 21:58

2 Answers2

1

Survival Analysis has been selected for the above problem as it provides the time when a particular event will occur. In the approach various models are under consideration and currently using coxph which we deem as the best model.

B K
  • 723
  • 8
  • 17
  • Can you explain how did you calculate "survival_in_days" for your case? I am building a similar case. – Learner Mar 08 '18 at 06:30
0

Machine learning might be a bit fancy of a term for this problem. I think you should just start with modeling this as a Poisson process. You can't really predict when something will happen, but you can predict what the odds are of the event happening before time X.

JoshK
  • 337
  • 4
  • 16
  • As a example, Survival analysis is generally defined as a set of methods for analyzing data where the outcome variable is the time until the occurrence of an event of interest. If my understanding is correct, this method will predict when particular event will occur. My doubt was which are the similar methods or models or approaches for the problem? – B K May 11 '16 at 17:29
  • Sorry, I just meant that it sounds like you have basic survival data, right? Then I would think that starting with trying to see if it fits a Poisson process. If you have some rich data then you would have something to dig into a bit more. – JoshK May 11 '16 at 18:04
  • If you have some background data then you can try the "survival" package in R. It has a few different models in it. – JoshK May 11 '16 at 18:05