0
$\begingroup$

Let me first say that I am a CS person and my knowledge about statistics is quite basic. I am trying to see what predictive analysis to use for a problem I am trying to solve. I will try to make my problem general also so that anyone can benefit from the question. Basically, I have a group of objects that I am watching over a certain period of time. By the end of that period, some objects remain unchanged, other objects are modified by a small amount, while other ones change completely. I am trying to look at a small history of these objects and predict how they will change in the future. I want to identify what are the parameters that affect their future behavior and how they affect their future behavior. I dont know what part of statistics should I look into. I think it is important to say that the outcome is discrete. I want to classify those objects into "stable", "change by a small value", or completely change.

$\endgroup$
3
  • $\begingroup$ You might have over-generalized your problem to the point where a good answer would have to cover most of statistics and data mining. Why don't you just ask the question you actually have? $\endgroup$ – whuber Oct 20 '14 at 21:34
  • 1
    $\begingroup$ This is almost my problem. Instead of objects, I actually have memory pages. Each memory page is 4K. In a period of five minutes, some of these pages will change completely. Others will stay the same. Others will change by a little bit. I want to classify pages into three categories. I have a second by second data of the whole memory for the whole five minutes. So I do know how the pages will change. I was wondering if I can look into a shorter amount of time. Lets say three seconds and be able to predict what category this page will fall into. $\endgroup$ – Keeto Oct 20 '14 at 21:39
  • $\begingroup$ Is there something special about these three categories? Or, if you could, would you prefer to predict how much the pages will change or when they change? In short, sharing some sense of why you want to do this will help generate useful advice. $\endgroup$ – whuber Oct 20 '14 at 22:23
0
$\begingroup$

The main portion of your question can be answered by : Investigate ARMA models which can be useful to analyze and predict a time series. Arbitrary classification is yet another matter. One would have to specify the classification rules and the use the forecasts vis-a-vis recent values in order to bucket the series into your three classes.

$\endgroup$

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.