2
$\begingroup$

I don't have much experience with ML but am looking for some guidance on where I should start.

I have timestamp data mapped to a certain category, e.g. travel, food, groceries, etc.

Judging by some other posts and answers, would it be correct to break the data into these intervals (day, week, month, year) and then try to detect patterns across the intervals? Are there any particular approaches or algorithms I should investigate first? Thanks!

$\endgroup$
  • $\begingroup$ I would recommend looking into Time Series Analysis. It is a general topic that has many approaches to extracting meaningful information from time series. $\endgroup$ – ryan Jun 22 '17 at 18:31
  • 1
    $\begingroup$ What are you trying to achieve? What's your goal? What do you want to learn from the data? What patterns are you looking for? $\endgroup$ – D.W. Jun 22 '17 at 20:21
  • 1
    $\begingroup$ Just trying to look at a series of transactions, and trying to see if any are repeated at a certain interval, such as buying coffee once a day, or a recurring bill each month, etc $\endgroup$ – holtc Jun 22 '17 at 21:50
  • 1
    $\begingroup$ @holtc: depending on how easy it is to uniquely identify a transaction type, what the accepted tolerance is in what constitutes an "interval" and whether you don't need to detect more complex patterns or interactions - you might achieve what you want with a relatively simple algorithm instead of bringing out the big guns $\endgroup$ – Amnon Jun 22 '17 at 22:53
0
$\begingroup$

Given an event type and a sequence of times at which it has occurred, there are several ways to check whether it seems to recur periodically.

One simple approach is to look at the interarrival times (the time between each occurrence and the previous occurrence), and histogram those; if it is periodic, you'll see a sharp peak in the histogram.

Another approach is to use autocorrelation to detect periodicity. See, e.g., https://en.wikipedia.org/wiki/Autocorrelation. The Fourier transform can be useful here.

$\endgroup$

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.