0
$\begingroup$

I've started to work on time series. I was wondering what would be the best data normalizing and pre-processing technique for non-linear models, specifically, neural networks.

One I can think of is min-max normalization

$$z = \frac{x - min(x)}{max(x) - min(x)}$$

$\endgroup$
2
  • $\begingroup$ are we assuming that the min and max do not change over time? specifically, from the training data to the test data. If there is trending or drift then z could end up negative or above 1 for large portions of test data using this normalisation method. $\endgroup$
    – Mike NZ
    Apr 19, 2021 at 6:20
  • $\begingroup$ @Leo would you share what neural network model using? At least similar popular model. $\endgroup$
    – Cloud Cho
    Dec 4, 2021 at 5:55

2 Answers 2

0
$\begingroup$

I would say any normalization such as min-max or standard deviation is fine as far as the scaling factor is provided as a feature, since time-series of different scale might behave differently.

$\endgroup$
0
$\begingroup$

I just finished my masters thesis project for multivariate time series prediction.

  1. The standard approach is to normalize the data using minmax or z-score (there is one research paper that found it really didn’t matter which normalization technique was used).

  2. You will most likely need to transform the dataset to a supervised problem (X, y)

  3. You will need to transform the data into a format used by most neural networks (samples, timesteps, feature)

How to Convert a Time Series to a Supervised Learning Problem in Python

Multivariate Time Series Forecasting with LSTMs in Keras

$\endgroup$

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .