How To Use and Remove Trend Information From Time Series Data in Python

How to Use and Remove Trend Information from Time Series Data in P... https://machinelearningmastery.
com/time-series-trends-in-python/
Navigation
Click to Take the FREE Time Series Crash-Course
Search...
How to Use and Remove Trend Information from

Time Series Data in Python
by Jason Brownlee on December 21, 2016 in Time Series
Tweet Share Share
Last Updated on August 21, 2019
Our time series dataset may contain a trend.
A trend is a continued increase or decrease in the series over time. There can be benefit in identifying,
modeling, and even removing trend information from your time series dataset.
In this tutorial, you will discover how to model and remove trend information from time series data in
Python.
After completing this tutorial, you will know:
The importance and types of trends that may exist in time series and how to identify them.
How to use a simple differencing method to remove a trend.
How to model a linear trend and remove it from a sales time series dataset.
Discover how to prepare and visualize time series data and develop autoregressive forecasting models
in my new book, with 28 step-by-step tutorials, and full python code.
Let’s get started.
Updated Apr/2019: Updated the link to dataset.
1 of 22 4/5/2020, 2:19 PM
How to Use and Remove Trend Information from Time Series Data in P... https://machinelearningmastery.com/time-series-trends-in-python/
How to Use and Remove Trend Information from Time Series Data in Python
Photo by john78727, some rights reserved.
Trends in Time Series

A trend is a long-term increase or decrease in the level of the time series.
In general, a systematic change in a time series that does not appear to be periodic is known
as a trend.
— Page 5, Introductory Time Series with R
Identifying and understanding trend information can aid in improving model performance; below are a
few reasons:
Faster Modeling: Perhaps the knowledge of a trend or lack of a trend can suggest methods and
make model selection and evaluation more efficient.
Simpler Problem: Perhaps we can correct or remove the trend to simplify modeling and improve
model performance.
More Data: Perhaps we can use trend information, directly or as a summary, to provide additional
information to the model and improve model performance.
Types of Trends
There are all kinds of trends.
Two general classes that we may think about are:
2 of 22 4/5/2020, 2:19 PM
Deterministic Trends: These are trends that consistently increase or decrease.

Stochastic Trends: These are trends that increase and decrease inconsistently.
In general, deterministic trends are easier to identify and remove, but the methods discussed in this
tutorial can still be useful for stochastic trends.
We can think about trends in terms of their scope of observations.
Global Trends: These are trends that apply to the whole time series.
Local Trends: These are trends that apply to parts or subsequences of a time series.
Generally, global trends are easier to identify and address.
Identifying a Trend
You can plot time series data to see if a trend is obvious or not.
The difficulty is that in practice, identifying a trend in a time series can be a subjective process. As such,
extracting or removing it from the time series can be just as subjective.
Create line plots of your data and inspect the plots for obvious trends.
Add linear and nonlinear trend lines to your plots and see if a trend is obvious.
Removing a Trend
A time series with a trend is called non-stationary.
An identified trend can be modeled. Once modeled, it can be removed from the time series dataset. This
is called detrending the time series.
If a dataset does not have a trend or we successfully remove the trend, the dataset is said to be trend
stationary.
Using Time Series Trends in Machine Learning

From a machine learning perspective, a trend in your data represents two
1. Remove Information: To remove systematic information that distorts

input and output variables.
2. Add Information: To add systematic information to improve the relati
output variables.
Specifically, a trend can be removed from your time series data (and data
preparation and cleaning exercise. This is common when using statistical
forecasting, but does not always improve results when using machine lear
3 of 22 4/5/2020, 2:19 PM
Alternately, a trend can be added, either directly or as a summary, as a new input variable to the
supervised learning problem to predict the output variable.
One or both approaches may be relevant for your time series forecasting problem and may be worth
investigating.
Next, let’s take a look at a dataset that has a trend.
Stop learning Time Series Forecasting the slow way!

Take my free 7-day email course and discover how to get started (with sample code).
Click to sign-up and also get a free PDF Ebook version of the course.
Shampoo Sales Dataset

This dataset describes the monthly number of sales of shampoo over a 3 year period.
The units are a sales count and there are 36 observations. The original dataset is credited to
Makridakis, Wheelwright, and Hyndman (1998).
Download the dataset.
Below is a sample of the first 5 rows of data, including the header row.
Below is a plot of the entire dataset, where you can learn more and downl
The dataset shows an increasing trend.
4 of 22 4/5/2020, 2:19 PM
Shampoo Sales Dataset
Load the Shampoo Sales Dataset

Download the dataset and place it in the current working directory with the filename “shampoo-
sales.csv“.
Download the dataset.
The dataset can be loaded with a custom date parsing routine as follows:
Running the example loads the dataset and creates a plot.
5 of 22 4/5/2020, 2:19 PM
Shampoo Sales Dataset Plot
Detrend by Differencing
Perhaps the simplest method to detrend a time series is by differencing.
Specifically, a new series is constructed where the value at the current time step is calculated as the
difference between the original observation and the observation at the previous time step.
This has the effect of removing a trend from a time series dataset.
We can create a new difference dataset in Python by implementing this dir

observations can be created.
Below is an example that creates the difference detrended version of the S
6 of 22 4/5/2020, 2:19 PM
Running the example creates the new detrended dataset and then plots the time series.
Because no difference value can be created for the first observation (there is nothing for it to be
subtracted from), the new dataset contains one less record. We can see that indeed the trend does
appear to have been removed.
Shampoo Sales Dataset Difference Detrende
This approach works well for data with a linear trend. If the trend is quadratic (the change in the trend
also increases or decreases), then a difference of the already-differenced
second level of differencing. This process can be further repeated if neede
7 of 22 4/5/2020, 2:19 PM
Because differencing only requires the observation at the previous time step, it can easily be applied to
unseen out-of-sample data to either preprocess or provide an additional input for supervised learning.
Next, we will look at fitting a model to describe the trend.
Detrend by Model Fitting

A trend is often easily visualized as a line through the observations.
Linear trends can be summarized by a linear model, and nonlinear trends may be best summarized
using a polynomial or other curve-fitting method.
Because of the subjective and domain-specific nature of identifying trends, this approach can help to
identify whether a trend is present. Even fitting a linear model to a trend that is clearly super-linear or
exponential can be helpful.
In addition to being used as a trend identification tool, these fit models can also be used to detrend a
time series.
For example, a linear model can be fit on the time index to predict the observation. This dataset would
look as follows:
The predictions from this model will form a straight line that can be taken as the trend line for the
dataset. These predictions can also be subtracted from the original time series to provide a detrended
version of the dataset.
The residuals from the fit of the model are a detrended form of the dataset. Polynomial curve fitting and
other nonlinear models can also be used.
We can implement this in Python by training a scikit-learn
8 of 22 4/5/2020, 2:19 PM
Running the example first fits the linear model to the integer-indexed observations and plots the trend
line (green) over the original dataset (blue).
Shampoo Sales Dataset Plot With Trend
Next, the trend is subtracted from the original dataset and the resulting de
9 of 22 4/5/2020, 2:19 PM
Shampoo Sales Dataset Model Detrended
Again, we can see that this approach has effectively detrended the dataset. There may be a parabola in
the residuals, suggesting that perhaps a polynomial fit may have done a better job.
Because the trend model takes only the integer index of the observation as input, it can be used on new
data to either detrend or provide a new input variable for the model.
Further Reading
Below are some additional resources on trend estimation and detrending i
Linear trend estimation on Wikipedia

Detrending Notes, GEOS 585A, Applied Time Series Analysis [PDF]
Update: download from this page.
Summary
In this tutorial, you discovered trends in time series data and how to remov
10 of 22 4/5/2020, 2:19 PM
Specifically, you learned:
About the importance of trend information in time series and how you may be able to use it in
machine learning.
How to use differencing to remove a trend from time series data.
How to model a linear trend and remove it from time series data.
Do you have any questions about detrending or about this tutorial?

Ask your questions in the comments below and I will do my best to answer.
Want to Develop Time Series Forecasts with Python?

Develop Your Own Forecasts in Minutes
...with just a few lines of python code
Discover how in my new Ebook:
Introduction to Time Series Forecasting With Python
It covers self-study tutorials and end-to-end projects on topics like: Loading

data, visualization, modeling, algorithm tuning, and much more...
Finally Bring Time Series Forecasting to

Your Own Projects
Skip the Academics. Just Results.
SEE WHAT'S INSIDE
Tweet Share Share
About Jason Brownlee

Jason Brownlee, PhD is a machine learning specialist who teaches developers how to get results
with modern machine learning methods via hands-on tutorials.
View all posts by Jason Brownlee →
How To Backtest Machine Learning Models for Time Series Forecasting

How to Identify and Remove Seasonality from Time Series Data with Python
11 of 22 4/5/2020, 2:19 PM
45 Responses to How to Use and Remove Trend Information from Time

Series Data in Python
REPLY
Carmen Mardiros January 27, 2017 at 11:40 pm #
Really useful Jason. One question. Would the de-trended values be used as additional features
or instead of the original features. Also, is it common to de-trend the target variable as well? If yes, do the
de-trended values become the new target variable? If yes, then what is the process for obtaining full
predictions from ML models (for sklearn at least, I’m guessing a custom transformer needs to store the
removed trend and add it back in using inverse_transform).
Thank you!
REPLY
Jason Brownlee January 28, 2017 at 7:49 am #
Hi Carmen,
The discussion in this post is for de-trending a univariate time series. If you wish to detrend a
univariate time series, you would treat it as a suite of univariate time series and de-trend one at a
time.
The whole series may be de-trended and is used as input for a learning algorithm like ARIMA. See
this post:
http://machinelearningmastery.com/arima-for-time-series-forecasting-with-python/
Alternatively, the detrended series may be used as an additional input for a machine learning
algorithm.
REPLY
Giovanni May 24, 2017 at 7:28 pm #
Very useful, thank you! I’m wondering if it is possible to give back th

detrended time series…do you know how I can do it? Thanks
Jason Brownlee June 2, 2017 at 11:32 am
Yes, if the trend was removed with subtraction, it can be adde
Gabriele February 3, 2020 at 9:22 pm #
12 of 22 4/5/2020, 2:19 PM
I guess you can just start with the initial value of the original timeseries, and then recreate
the original timeseries by adding each delta obtained with the differentiation.
REPLY
Marianico July 21, 2017 at 10:52 pm #
Would detrending technique improve ARIMA forecasting?
REPLY
Jason Brownlee July 22, 2017 at 8:35 am #
It can. Try it and see. ARIMA has it built in with the q var.
REPLY
Ahmed August 11, 2017 at 12:55 pm #
Great tutorial! really thanks for such useful information. My question is, Is it possible to have
linear and non-linear trend added to the same data set? If yes, Shall I remove the linear part first, then
the non-linear part (which is a straight line fitted to the data and then subtract it) and then remove the
non-linear part (fit a low-order polynomial to the signal and then subtract it)? Or just removing the non-
linear part would be sufficient ??
REPLY
Jason Brownlee August 12, 2017 at 6:46 am #
Interesting question. Generally, I would recommend trying both and see which model results
in better forecasts.
REPLY
Weber August 12, 2017 at 9:50 pm #
Suppose the time series shows downward trend most of the times
previous time stamp may result in a negative value. Perhaps simple moving averages can be used in this
case..?
Jason Brownlee August 13, 2017 at 9:53 am
Perhaps. But what is wrong with negative values – you’re wor

You can always reverse the transform on the prediction to get back to t
13 of 22 4/5/2020, 2:19 PM
REPLY
Weber August 13, 2017 at 7:00 pm #
Thanks for your reply. This technique would fail for Poisson Auto regression model. Do
you have any suggestions to detrending in event count models?
REPLY
Not off-hand, sorry Weber.
REPLY
Mamta August 21, 2017 at 4:35 am #
Dear Jason,
I followed your approach to detrend the time series data before fitting a neural network model on the
training set.
I need to forecast hourly values for next day.
When I have to use my trained model to predict on unseen data, how do I deal with the trend ?
Do I have to predict the trend along with the the usual observations, as my model is trained on detrended
data? Should this predicted trend be added to the predicted output of the trained model ?
Thanks
Mamta
REPLY
You can remove a trend by differencing. You can then invert the differencing after you have
made a prediction.
See here:
http://machinelearningmastery.com/remove-trends-seasonality-difference
and here:
http://machinelearningmastery.com/difference-time-series-dataset-pyth
Tushar February 8, 2018 at 11:14 pm #
Dear Jeson
Sir, I have a time series data obtained from integration of another data. The
exponentially decaying in amplitude with time. Would you please suggest m
14 of 22 4/5/2020, 2:19 PM
(identify) linear or non-linear trend (Equation of poly nomial) present in the derived time series data (i.e.
obtained from integration of another data)?
Thanks
REPLY
Jason Brownlee February 9, 2018 at 9:09 am #
I don’t have an example. Perhaps you can use a power transform on the obs to stabilize the
changing variance, then try removing the cycle?
Or perhaps you can use an scipy function call to model the exponential trend and remove it
manually?
Perhaps explore a few approaches and see what looks good.
REPLY
Faiza March 16, 2018 at 12:58 am #
Dear Jason
Sir plzz give me any example i have a data of time series in 4 quarters.how i solve this “assuming the
multiplicative model and using the movig average method,obtain the detrended time seies”??????
REPLY
Jason Brownlee March 16, 2018 at 6:20 am #
Sorry, I cannot write code for you. Perhaps hire a contractor?
REPLY
Rahul May 21, 2018 at 7:28 pm #
Hi Jason,
Really useful tutorial. Just wanted to inform you that Detrending Notes, GEOS 585A, Applied Time Series
Analysis [PDF] in Further Reading section is not working. If possible, could
Regards,
Rahul Sathyajit
Jason Brownlee May 22, 2018 at 6:27 am
Thanks, I’ve added a link to the page to learn more.
15 of 22 4/5/2020, 2:19 PM
REPLY
hamid August 16, 2018 at 8:47 pm #
Jason,
Thank you for your incredible post.
So, in one of your LSTM posts, you used multivariate weather input features (x1,x2,x3,x4) to predict
another weather parameter (Y). Should I also use detrending techniques for the input features which also
have trends ( for example, suppose x1 and x4 have trend) ? If yes, I don’t feel confident about my results,
as I’ve changed my input data to a new one. If no, maybe my model cannot predict well. What am I
missing?
REPLY
Ideally yes. LSTMs are sensitive to trends and seasonality.
REPLY
HAMID August 21, 2018 at 6:31 am #
Hello Jason
What is the best way to select the optimal subset(rows) to be trained by LSTM?
regards
Salah
REPLY
Careful and systematic experimentation.
REPLY
Jakub Kajan October 18, 2018 at 3:59 am #
Hello Jason.
Thank you very much for great tutorial. I tried to apply diff to data via pandas diff, but with diff data model
was not able to converge, I tried different topologies, neurons in layers etc.
But when I don’t modify data, fit is really good. Have you ever meet such issue, or could you know some
good reading related to this topic?
Regards Jakub.
Jason Brownlee October 18, 2018 at 6:37 am
16 of 22 4/5/2020, 2:19 PM
Yes, MLPs and CNNs seem to perform well even better when the data is non-stationary, at
least in a few tests I performed.
REPLY
Theekshana November 7, 2018 at 7:38 pm #
Hi Janson,
Is there any well-known method for identifying the time series trend. As an example given a time series,
the method should be able to give a value that says the series increasing, decreasing or stabilized (the
value should represent the trend perfectly. some kind of a scale maybe 0-100, 0 for decreasing 100 for
increasing at a higher rate, 50 for stable etc).
From your example, I can see that the fitted line can be used to come up with some value about the trend
( maybe using the gradient ). Is there any sophisticated way to do this?
Thank you.
Great tutorial. Big fan of this site…
REPLY
Jason Brownlee November 8, 2018 at 6:06 am #
Yes, the gradient, rise/run.
REPLY
S February 2, 2019 at 9:52 pm #
thanks for the tutorial!
Question – once the trend has been removed (as you have in your last image), one could use that
detrended data to make some predictions with ARIMA. How would one interpret the predicted value? Let
us assume it goes up by 2 ‘units’ in the next period = could expect an increase of 2 x the scale that was
used in the original data set?
Jason Brownlee February 3, 2019 at 6:16 am
The trend can be added back to any predictions made and the interpretation would then be
specific to the domain.
Afreen February 14, 2019 at 9:47 pm #
Hi Jason,
17 of 22 4/5/2020, 2:19 PM
Thank you for this tutorial!

So, I am working on a problem where I need to incorporate time-independent features to make
predictions for target variable that is time dependent.
I detrend the target (time-dependent) and I build a supervised ML model with features as the time-
independent features. To the prediction of this model, I add back the trend.
Would this be the right way to go about it?
Does any of your other tutorial talk about it? Searched for it but couldn’t find it.
Thanks a lot for your help!
REPLY
Jason Brownlee February 15, 2019 at 8:05 am #
Perhaps add the new variables as exog variables for a time step.
Otherwise, perhaps use an ensemble model that combined predictions from a model on the time
series and a model no the static data?
REPLY
Leen March 16, 2019 at 10:59 pm #
Hello Jason,
Thank you for the awesome tutorial!!
I wanted to ask if we are to detrend and deseasonalize, van’t we just use the seasonal_decompose()
function, and extract the trend and seasonality components from it and leave the residual ?? Is the
residual considered detrended and deseasonalized ?
Example
result = seasonal_decompose(self.series, model=’additive’, freq=frequency)
residual = result.resid # this is what I am reffering to.
Thanks in advance
Jason Brownlee March 17, 2019 at 6:22 am
Perhaps. It may not be reliable for your data. It is better to hav

understand the structure of each.
Leen March 17, 2019 at 6:48 am #
So which is more reliable ? detrending and deseasonalizi
18 of 22 4/5/2020, 2:19 PM
from seasonal_decompose()
REPLY
It depends on your project, in general I recommend detrending and seasonal

adjustment prior to modeling or as part of modeling in the case of methods like SARIMA.
REPLY
deepika March 20, 2019 at 10:15 pm #
Thanks for this post, Jason!

I was wondering, if I am performing an LSTM with multiple variables, and I observe a trend only for 2 out
of the 5 variables,
1. do I detrend all 5 variables by differencing?
2. If one of them is categorical and I have coded it as(1,-1) does this have to be differnced as well?
Thanks a lot!
REPLY
Try modeling the data directly, then with differencing on the series with trends to see if there
is an impact.
For categorical vars, explore integer encode, one hot encode and perhaps embeddings.
REPLY
Teyang July 8, 2019 at 6:22 pm #
Hi Jason,
Great post!
Can I check with you regarding the model fitting method.
When you mentioned subtracting the trend from the original data, is it the s
of the linear/quadratic/polynomial model?
Thanks!
19 of 22 4/5/2020, 2:19 PM
REPLY
Jason Brownlee July 9, 2019 at 8:06 am #
Similar.
REPLY
Laura August 23, 2019 at 12:38 am #
Hi Jason,
If the goal is to forecast the future N timesteps of a timeseries showing a stochastic trend, then we should
fit a model to approximate the trend, remove this approximated trend from the original data and use
models such as ARIMA to predict there residuals. However, the final forecast will be Arima forecast +
trend forecast.
The timeseries data I work with is not well approximated by a linear regression, it consists of random
patterns with the trend going up and down at different amplitudes (like a random walk). I didn’t find yet a
model which can approximated it well enough. What model do you recommend using for this kind of
stochastic trend approximation?
Thanks!
REPLY
Typically an ARIMA can remove the trend for you via the d parameter.
I recommend testing a suite of linear and non-linear models in order to discover what works best for
your specific dataset:
https://machinelearningmastery.com/how-to-develop-a-skilful-time-series-forecasting-model/
REPLY
Ha My Nguyen February 29, 2020 at 5:42 pm #
Hi, I run the code on Python and I dont know why this function kept showing error
datetime.strptime. Is that possible to write in another way?
Jason Brownlee March 1, 2020 at 5:24 am
Sorry to hear that, perhaps this will help:

https://machinelearningmastery.com/faq/single-faq/why-does-the-code-in-the-tutorial-not-work-for-me
20 of 22 4/5/2020, 2:19 PM
Leave a Reply
Name (required)
Email (will not be published) (required)
Website
SUBMIT COMMENT
Welcome!
My name is Jason Brownlee PhD, and I help developers get results with machine learning.
Read more
Never miss a tutorial:
Picked for you:
How to Create an ARIMA Model for Time Series Forecasting in Python
How to Convert a Time Series to a Supervised Learning Problem in Pyt
Time Series Forecasting as Supervised Learning
21 of 22 4/5/2020, 2:19 PM
11 Classical Time Series Forecasting Methods in Python (Cheat Sheet)
How To Backtest Machine Learning Models for Time Series Forecasting
Loving the Tutorials?
The Time Series with Python EBook

is where I keep the Really Good stuff.
SEE WHAT'S INSIDE
© 2019 Machine Learning Mastery Pty. Ltd. All Rights Reserved.

Address: PO Box 206, Vermont Victoria 3133, Australia. | ACN: 626 223 336.
LinkedIn | Twitter | Facebook | Newsletter | RSS
Privacy | Disclaimer | Terms | Contact | Sitemap | Search
22 of 22 4/5/2020, 2:19 PM

How To Use and Remove Trend Information From Time Series Data in Python

Uploaded by

Copyright:

Available Formats

You might also like

How To Use and Remove Trend Information From Time Series Data in Python

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

How To Use and Remove Trend Information From Time Series Data in Python

Uploaded by

Copyright:

Available Formats

How to Use and Remove Trend Information from Time Series Data in P... https://machinelearningmastery.

Click to Take the FREE Time Series Crash-Course

How to Use and Remove Trend Information from

Tweet Share Share

Last Updated on August 21, 2019

Our time series dataset may contain a trend.

After completing this tutorial, you will know:

Let’s get started.

Updated Apr/2019: Updated the link to dataset.

Trends in Time Series

— Page 5, Introductory Time Series with R

Two general classes that we may think about are:

Deterministic Trends: These are trends that consistently increase or decrease.

We can think about trends in terms of their scope of observations.

Generally, global trends are easier to identify and address.

Using Time Series Trends in Machine Learning

1. Remove Information: To remove systematic information that distorts

Next, let’s take a look at a dataset that has a trend.

Stop learning Time Series Forecasting the slow way!

Shampoo Sales Dataset

Download the dataset.

The dataset shows an increasing trend.

Shampoo Sales Dataset

Load the Shampoo Sales Dataset

Download the dataset.

Running the example loads the dataset and creates a plot.

Shampoo Sales Dataset Plot

We can create a new difference dataset in Python by implementing this dir

Below is an example that creates the difference detrended version of the S

Shampoo Sales Dataset Difference Detrende

Next, we will look at fitting a model to describe the trend.

Detrend by Model Fitting

We can implement this in Python by training a scikit-learn

Shampoo Sales Dataset Plot With Trend

Shampoo Sales Dataset Model Detrended

Linear trend estimation on Wikipedia

Specifically, you learned:

Do you have any questions about detrending or about this tutorial?

Want to Develop Time Series Forecasts with Python?

It covers self-study tutorials and end-to-end projects on topics like: Loading

Finally Bring Time Series Forecasting to

SEE WHAT'S INSIDE

Tweet Share Share

About Jason Brownlee

How To Backtest Machine Learning Models for Time Series Forecasting

45 Responses to How to Use and Remove Trend Information from Time

Very useful, thank you! I’m wondering if it is possible to give back th

Jason Brownlee June 2, 2017 at 11:32 am

Yes, if the trend was removed with subtraction, it can be adde

Gabriele February 3, 2020 at 9:22 pm #

Would detrending technique improve ARIMA forecasting?

Jason Brownlee August 13, 2017 at 9:53 am

Perhaps. But what is wrong with negative values – you’re wor

Not off-hand, sorry Weber.

I need to forecast hourly values for next day.

Tushar February 8, 2018 at 11:14 pm #

Perhaps explore a few approaches and see what looks good.

Sorry, I cannot write code for you. Perhaps hire a contractor?