4

I am thinking of making a model or whatever name i don't know. Idea is this that I suffers from headache, say twice in a month. What i want to do is that i want to make some kind of system in which i can predict my next headache date based on previous dates. i thought of making a graph and then to observe some kind of uniformity in it so as to predict. It can have points like (2,22) meaning headache on 22nd of February. Also this thing makes sense only if i keep my routine same in every aspect.Now i want to make this system more rigorous but i donot have tools for that. Can someone suggest if anything serious can be done on this direction?

Gathdi
  • 1,402
  • Interesting question. There are many models which might fit this, but one can only decide after having seen the data. How much dates have you recorded? If you post them, we can help looking for some kind of relationship. – Bib-lost Jun 11 '16 at 14:56
  • A first approximation might be an exponential distribution, in which case because of the memorylessness property the date of the current headache may give very little additional information about the date of the next one. – André Nicolas Jun 11 '16 at 15:02
  • 1
    I think the event of getting a headache is independent on the number of days in a month, but rather the number of days that have passed after the last headache. Therefore I suggest keeping track of the number of days between consecutive headaches. I think you will find 2 peaks. One peak will occur at small values (1, 2, 3), because headaches may last a few days. The other peak will be a bit further off (10~30). The last peak will most likely be where you're interested in, so you can analyze that! I expect you can fit a normal distribution to that (by the law of large numbers). – Jasper Jun 11 '16 at 15:07
  • i will update with data – Gathdi Jun 11 '16 at 15:10

1 Answers1

1

You might be able to use some technique from regression analysis to model the headache occurences. For example, you could use the method of least squares to predict the number of headaches per [time period], if it turns out that they occur in some sort of cycle.

For example, let's say that you have three headaches in week one, two in week 2, five in week three, and zero in week four. This isn't a lot of data, but you could still represent it by the points $$(1,3),\quad(2,2),\quad(3,5),\quad(4,0)$$ Now perform a least squares analysis on it to come up with a polynomial fit. In reality, any form of curve fitting may come up with a good answer, provided that you have enough data and there really is a pattern.

HDE 226868
  • 2,354
  • This certainly might work, but it makes a lot of assumptions which might not be justified untill we see the data. For example: why would day of the month correlate with month? Are consequent headaches independent, or does one headache influence the probability of another one happening shortly thereafter? And so on ... – Bib-lost Jun 11 '16 at 15:01
  • @Bib-lost I know; it's a start. – HDE 226868 Jun 11 '16 at 15:02
  • i know no regression analysis – Gathdi Jun 11 '16 at 15:08