Dear Basri Badyalina

These blogs and link might help, have a look:

1. https://towardsdatascience.com/a-quick-deep-learning-recipe-time-series-forecasting-with-keras-in-python-f759923ba64

2. https://stats.stackexchange.com/questions/141503/neural-network-for-time-series-forecasting-single-input-single-output-theoretic

3. https://machinelearningmastery.com/how-to-develop-convolutional-neural-network-models-for-time-series-forecasting/

Kind Regards

Qamar Ul Islam

Qamar Ul Islam Thank You.

Dear Basri Badyalina ,

I suggest a kind of decomposition approach. First start with as much data as you have (or can) to train the network than try to reduce the data until you reach a point where the result is still acceptable. Of course the concrete answer highly depends on the domain.

Kind regards,

László

Dear László Csépányi-Fürjes ,

Thank you for your response. Is it possible to use decomposition approach in the initial step to determine the number of inputs for univariate data?. If I am not mistaken, the decomposition approach only starts after deciding the number of input variables. Correct me if I am wrong.

Regards

Basri B

Pleasure Basri Badyalina

Kind Regards

There are no specific "guidelines", if your are working with Python you can read/import your variables directly from a .csv file and after that you can normalize your data. It much depends on the task you're performing; if you are working with financial data you may need to perform an Augmented Dickey-Fuller test after doing the normalization.

https://scikit-learn.org/stable/auto_examples/preprocessing/plot_scaling_importance.html#sphx-glr-auto-examples-preprocessing-plot-scaling-importance-py

https://www.statsmodels.org/dev/generated/statsmodels.tsa.stattools.adfuller.html

Enrique Gonzalez N Thank You

Dear Basri Badyalina ,

I agree with Enrique Gonzalez N. It very much depends on your domain. And to answer your question, yes. You have to make a decision at the beginning for yourself. I suggest to try to find a paper where a similar data set was processed and use the same input variables as a starting point.

Regards,

László

László Csépányi-Fürjes Thank You. My case study right now using streamflow data a univariate times series data. Do you have any suggestions on how to decide on which input should I start?

So you are talking about a river... A typical univariate data is the water level. So maybe you could build a NN to predict the water level based on previous time series. This seems to be a useful paper for you: Article Water Level Prediction with Artificial Neural Network Models

László Csépányi-Fürjes Thank You.