4
$\begingroup$

I have a sensor that can detect minute changes in distance. It produces a time series.

I would like to point it at people and detect things like their sleeping pattern. How would one build a system that figures out what 'normal' means for this user and can say things like 'over the past 3 months your sleeping pattern has shifted 3 hours, this is bad'.

Here is a picture of what the series looks like over the course of one day

http://i.stack.imgur.com/9DbJy.jpg

One transformation of it we call position, it tries to illustrate the position the subject holds (laying down, sitting up, etc)

http://i.stack.imgur.com/h3G3j.png

Any and all ideas are welcome. I've skimmed "Time Series Analysis and Its Applications: With R Examples" by Shumway and Stoffer. I think I should explore Spectral Analysis and Filtering due to the cyclical nature of the data. I've been using python/pandas and am looking into doing work with statsmodels and scikit-learn.

EDIT1: Activity is the 'raw' data from the sensor. It is distance (higher values are closer to the sensor) but it's value lies in it's sensitivity. I want to detect breathing patterns if possible. As far as position, I'm not actually sure what the formula for that transformation was, I'll ask the author.

$\endgroup$
  • $\begingroup$ What are the numbers on the vertical axes? Codes for categories? Actual "distances"? Counts of something? Differences between two such values (as in the lower plot)? (Those four would suggest four different kinds of analysis.) $\endgroup$ – whuber Mar 28 '13 at 19:32

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.