2
$\begingroup$

I have time-series for creatinine levels in patients, which has missing samples, due to patients' irregular visits to doctors. The figure below represents the time-series for a patient.

Time-series data for a patient's Creatinine levels

Task: I need to interpolate the missing sample values, for which I can use other auxiliary data about the patient such as their age, gender, smoking habits, etc.

What would be a good direction to approach this task?

Improve this question
$\endgroup$
5
  • $\begingroup$ Please specify why you want to interpolate the data. This may strongly influence how you interpolate and whether interpolating is a good idea in the first place. $\endgroup$
    – Wrzlprmft
    Commented Nov 15, 2015 at 10:48
  • $\begingroup$ My eventual goal is to form clusters of patients that have similar creatinine progression patterns. But since the time-series have these missing data points, I can't use a similarity measure due to the unequal length series. If I do the interpolation I can use any similarity measure and clustering algorithm. $\endgroup$
    – Ankit Kapur
    Commented Nov 15, 2015 at 10:55
  • $\begingroup$ Sounds like you should rather be looking for a similarity measure that can deal with missing data points (which sounds like a good question for this site, unless already asked). Interpolating my introduce several biases to a similarity measure aimed at “regular” time series. To give a very simple example, depending on the interpolation method, time series with a relatively high amount missing data may be considered similar to each other due to being relatively smooth. $\endgroup$
    – Wrzlprmft
    Commented Nov 15, 2015 at 11:00
  • $\begingroup$ You're right that can be another way of doing it - Are you aware of any such similarity measures? $\endgroup$
    – Ankit Kapur
    Commented Nov 15, 2015 at 18:29
  • $\begingroup$ No, but as I said, this sounds like a good question for this site. $\endgroup$
    – Wrzlprmft
    Commented Nov 15, 2015 at 19:45

0

Reset to default

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.