0
$\begingroup$

I have multivariate time series data consists of 4 independent variables, 1 dependent variable (target variable), and spatial data (latitude and longitude). The data is taken from 5 different cities, therefore there are 5 pairs latitude and longitude (each pair represent each city). I have read several references that Conv2D can be applied on spatial cases. Therefore, I want to perform spatial analysis using Conv2D, but my data is not image (I have read that Conv2D commonly used image data as input). Then, the temporal analysis is perform using LSTM (I want to use both method to perform spatial (using Conv2D) and temporal (LSTM) analysis).

My questions are:

  1. Can I use Conv 2D to perform spatial analysis on my data? (There are spatial heterogeneity in the data, i.e. each city has different characteristics from the other city)
  2. If I can use Conv2D, what is the appropriate input shape? (My current data shape is (samples, timesteps, features, locations) which is (500, 24, 4, 5))

What I use:

  • Python
  • Keras with tensorflow

Any help in this regard will be highly appreciated. Thank you!

Improve this question
$\endgroup$

0

Reset to default

Your Answer

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