2
$\begingroup$

I have a dataset containing timestamp and temperature. For each day, I have 1440 values viz., I have data for every minute of that day(60minutes * 24hrs = 1440).

The Dataset looks like this:

enter image description here

As an initial step, I gathered day1 data to predict day2 data. I have tried AR, ARIMA, SARIMAX models but I didn't find any positive results. I think this is multivariate since the time and the temperature values changes with respect to date. I need guidance to choose the ML model that will suit for my dataset and it should be able to predict next day/ next month

Improve this question
$\endgroup$

1 Answer 1

Reset to default
1
$\begingroup$

Just for clarification: your description (1 sample per minute) does not match the example data (far fewer data points which is understandable, but also two data points in one minute which contradicts the initial assertion.) If your actual measurements are like that you should first work on the sampling process to get reliable data.

For creating predictions, you need to have a reasonable model of the observed process. If you're measuring environmental temperatures, you will basically have three causes of variation:

  1. A day/night cycle
  2. A seasonal (summer/winter) cycle
  3. Local weather fluctuation

From only one day of samples, the only thing you can reasonably predict is that the next day will look mostly the same. If you collect more data over a year, you will be able to extract a seasonal cycle and estimate the deviations caused by local weather. "You" means either you as a researcher or any machine learning system that you program. Without sufficient historical data it is impossible to make good predictions (and even with sufficient data it's hard.)

Improve this answer
$\endgroup$
4
  • $\begingroup$ I have edited the sample data image for your reference. I have three causes as startup, steady and start down. I have tried seasonal ARIMA wherein the predictions mimic the raw data itself(very minor changes in the values) $\endgroup$
    – Koushik Raghav S
    Commented Jan 27, 2020 at 12:13
  • $\begingroup$ Thank you, this sample makes much more sense. However, without knowing what the actual process is that you're measuring, it is still impossible to come up with a reasonable model. The numbers don't look like typical outside temperatures... $\endgroup$
    – Hans-Martin Mosner
    Commented Jan 27, 2020 at 12:20
  • $\begingroup$ I have 60,000(approx) temperature data(for a month with 1minute interval time) with respect to the timestamp. I have to predict the next 60,000. Pls suggest me which model can fit for this dataset to predict. $\endgroup$
    – Koushik Raghav S
    Commented Jan 27, 2020 at 12:40
  • $\begingroup$ Sorry, I can't suggest a specific model as the information that you give is still really sparse. My intention was to point at the issues your data might have that would prevent any ML algorithm from achieving reasonable predictions. I'd probably try to tackle this problem with a RNN but that's just because I find the concept fascinating and RNNs seem to be a good fit for time-sequence data. But other techniques might be just as good or even better. $\endgroup$
    – Hans-Martin Mosner
    Commented Jan 27, 2020 at 16:41

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .