Steps to follow before feeding a time series dataset in to neural network in R

Question

Hi I'm having a time series dataset as below.

When it is plotted this is what I recognized.

I'm asked to forecast time series in this dataset. But I'm not sure on what are the preprocessing steps I need to be done, before feeding it to NN. Do I need to to input stationary data? Because in other models like ARIMA, SAMIRMA there are such steps.

1. How the data need to be preprocessed?
2. Can neuralnet R function for MLP used for this, because I was asked to use it?
3. When the pattern is seen, there are seasonal effects. How to use them.

Kindly help. What I need is a basic idea on areas that I need to focus on and what are the steps that I need to follow. Looking forward for your kind ideas as much as possible

Answer 1

if you are interested in general time-series forecasting in R, Rob Hyndman is the way to go. For his neural networks chapter, see here

fit <- nnetar(sunspotarea, lambda=0)
autoplot(forecast(fit,h=30))

For a more in-depth analysis of the Neural Network capabilities in R with the Keras / tensorflow packages, Francois Chollet's book should be the best source.

I know for a fact that it contains a chapter on time-series forecasting using RNN's.

Finally, h2o.ai is a good resource as well, they have an R package and this is some documentation on how to approach time series forecasting. Not sure if they did it specifically with NN's but their package definitely has this capability.

See the h2o package doc. For example, their h2o.deeplearning function builds a Deep Neural Network model using CPUs.