-1
$\begingroup$

I'm writing a seminar paper about the topic of forecasting NO2 (h=2) data. My dataset includes 130 timeseries (balanced panel) from 2010 to 2012. At the moment I'm struggling to find a suitable way to visualize that large data set in R. My goal at this stage of my paper is to do explorative/descriptive monitoring of those series (ACF/PACF, time series plots and component plost). For example I try to examine if characteritics of the time series change over time.

Already, I read the "Forecasting: Principles and Practice" book by Rob J Hyndman and George Athanasopoulos and Roger D. Peng's paper "A Method for Visualizing Multivariate Time Series Data". Unfortunately I still need your advice on the topic.

Improve this question
$\endgroup$
2
  • 1
    $\begingroup$ You might do better to narrow this down by editing. At the moment it is very broad. $\endgroup$
    – mdewey
    Commented May 5, 2018 at 14:09
  • 4
    $\begingroup$ You don't visualize 130 time series, raw data, all at once for a presentation of ideas. Instead, you break it down and focus on the relevant abstracted information. You provide raw data in a database for people that wish to explore it more in-depth. $\endgroup$
    – Sextus Empiricus
    Commented May 5, 2018 at 18:51

1 Answer 1

Reset to default
6
$\begingroup$

Rob Hyndman's presentation on "Visualization of big time series data" might give you some ideas for how to proceed: https://robjhyndman.com/seminars/big-time-series/. In his proposed approach, Rob Hyndman constructs a vector of features for each time series, where the features measure characteristics of that series: lag correlation, strength of seasonality, spectral entropy, etc. Then he uses a principal component decomposition on the features, and plots the first few principal components so that he can explore a lower dimensional space and discover interesting structure and unusual observations. See also: https://github.com/robjhyndman/anomalous.

Roger Peng's article on A Method for Visualizing Multivariate Time Series Data might be useful too: https://www.jstatsoft.org/article/view/v025c01/v25c01.pdf.

See also: https://arxiv.org/pdf/1708.07942.pdf.

Improve this answer
$\endgroup$

Your Answer

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

Not the answer you're looking for? Browse other questions tagged or ask your own question.