## Multivariate non-time series

Extensions of the recurrence plot and network approach to multivariate and spatial data.
David01
Junior
Posts: 1
Joined: Wed Feb 15, 2023 06:52
Affiliation (Univ., Inst., Dept.): University of Illinois at Chicago
Location: Chicago, IL, USA
Research field: Computer science

### Multivariate non-time series

Similar to this previous question: (http://forum.recurrence-plot.tk/viewtop ... f=4&t=3591).

I'd like to take multivariate data (independent feature1, feature2, ..., featureN) and convert each row into an image for classification. There are perhaps 10 features as well as a dependent variable, while the dependent variable is kept out of the image. A post elsewhere online suggested using recurrence plots to convert the row into an image. I tried doing this with the data (see attachment) using for example this Python library (https://github.com/laszukdawid/recurrence-plot).

A sample row of the dataset looks like this, where the exceedance column holds the dependent variable:
year month day dayofyear tide tide_gtm dtide_1 dtide_2 PrecipSum6 Precip24 lograin3T wet3 lograin7T wet7 Wtemp_B rad solar_noon WDIR WSPD awind owind exceedance
2016 6 1 153 176.91 1.00 -0.01 -0.09 0.00 0.00 -1.52 0.00 0.35 1.00 19.90 0.00 1.00 0.00 0.00 -0.00 0.00 1
Exceedance is a binary variable (0 or 1). My plan is to use existing image classification approaches to classify the images. Does the use of recurrence plots in this case appear valid? Is this approach too far out? Perhaps another multivariate-to-image approach is warranted?
Attachments
index.png (3.17 KiB) Viewed 57468 times
Norbert
Expert
Posts: 196
Joined: Wed Jan 4, 2006 11:03
Affiliation (Univ., Inst., Dept.): Potsdam Institute for Climate Impact Research, Germany
Location: Potsdam, Germany
Location: Potsdam Institute for Climate Impact Research, Germany

### Re: Multivariate non-time series

David01 wrote: Wed Feb 15, 2023 07:04 Similar to this previous question: (http://forum.recurrence-plot.tk/viewtop ... f=4&t=3591).

I'd like to take multivariate data (independent feature1, feature2, ..., featureN) and convert each row into an image for classification. There are perhaps 10 features as well as a dependent variable, while the dependent variable is kept out of the image. A post elsewhere online suggested using recurrence plots to convert the row into an image. I tried doing this with the data (see attachment) using for example this Python library (https://github.com/laszukdawid/recurrence-plot).

A sample row of the dataset looks like this, where the exceedance column holds the dependent variable:
year month day dayofyear tide tide_gtm dtide_1 dtide_2 PrecipSum6 Precip24 lograin3T wet3 lograin7T wet7 Wtemp_B rad solar_noon WDIR WSPD awind owind exceedance
2016 6 1 153 176.91 1.00 -0.01 -0.09 0.00 0.00 -1.52 0.00 0.35 1.00 19.90 0.00 1.00 0.00 0.00 -0.00 0.00 1
Exceedance is a binary variable (0 or 1). My plan is to use existing image classification approaches to classify the images. Does the use of recurrence plots in this case appear valid? Is this approach too far out? Perhaps another multivariate-to-image approach is warranted?
Yes, should work. In general, multivariate time series can be used to construct a phase space trajectory (each variable would correspond to one dimension). This is a common approach since many years. In you case, you would use this as a tool for ML based image classification where the image is simply the RP. There are already many publications on this topic.