Scribd
Upload
5 views

Data Normalizationand Standardization ATechnical Report

This technical report discusses two common data preprocessing techniques: normalization and standardization. Normalization rescales data to a specific range like 0 to 1, while standardization rescales data to have a mean of 0 and standard deviation of 1. The report provides formulas for performing normalization to different ranges and its inverse, denormalization. It also discusses z-score standardization and provides an example.

Uploaded by

Sergio Alumenda
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
5 views

Data Normalizationand Standardization ATechnical Report

This technical report discusses two common data preprocessing techniques: normalization and standardization. Normalization rescales data to a specific range like 0 to 1, while standardization rescales data to have a mean of 0 and standard deviation of 1. The report provides formulas for performing normalization to different ranges and its inverse, denormalization. It also discusses z-score standardization and provides an example.

Uploaded by

Sergio Alumenda
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 6

Machine Learning Technical Reports (2014) 1(1): 1-6

Data Normalization and Standardization: A Technical


Report
Peshawa Jamal Muhammad Ali*, and Rezhna Hassan Faraj
The Machine Learning Lab. at Koya University
Koya, Erbil, Iraq.
[email protected]

Cite the technical report:


Peshawa J. Muhammad Ali, Rezhna H. Faraj; “​Data Normalization and Standardization: A Technical
Report”​ , Machine Learning Technical Reports, 2014, 1(1), pp 1-6.
https://docs.google.com/document/d/1x0A1nUz1WWtMCZb5oVzF0SVMY7a_58KQulqQVT8LaVA/edit#

Information about the publisher:


Machine Learning Technical Reports is a periodical technical report published by the Machine Learning Lab. at Koya
University
Koya University, Building of Faculty of Engineering, KOY45
Koya 44023, Erbil, F.R. of Iraq
Contact: +9647707578801/+9647501138655, email: [email protected]

Abstract
This paper aims to clarify how and why data are normalized or standardized, these two
processes are used in the data preprocessing stage in which the data is prepared to be
processed later by one of the data mining and machine learning techniques like support vector
machine, neural network, etc. The two methods try to scale the data set. These two processes
are helpful in some cases and necessary in some other cases, most of the data mining and
machine learning tools include these two preprocessing techniques like in Weka or in Matlab.
This paper will simply define and present the use of these two data preprocessing techniques.

Normalization
It’s the process of casting the data to the specific range, like between 0 and 1 or between -1 and
+1. Normalization is required when there are big differences in the ranges of different features.
This scaling method is useful when the data set does not contain outliers. The theoretical
background of normalization can be easily understood from Figure (1). If it is required to cast
the data to the range 0,1 then:

Machine Learning Technical Reports is a periodical technical report published by the Machine Learning Lab. at Koya
University
Koya University, Building of Faculty of Engineering, KOY45
Koya 44023, Erbil, F.R. of Iraq
Contact: +9647707578801/+9647501138655, email: [email protected]
From Trigonometry:

valueAf terN ormalization − 0 valueBef oreN ormalization − min


1−0 = max − min
valueAf terN ormalization valueBef oreN ormalization − min
1 = max − min

valueBef oreN ormalization − min


v alueAf terN ormalization = max − min

x − min
or x′ = max − min

Denormalization
This process should be done if normalization applied. For example, to denormalize the a data
from the range 0, 1 below equation can be used:

x = [x′ * (max − min)] + min

where x’ is the normalized data and x is denormalized data, min and max are the same values
used previously in the normalization process.

2
To normalize the data to the range -1, +1 see Fig(2):

valueAf terN ormalization − (−1) valueBef oreN ormalization − min


1 − (−1) = max − min

valueAf terN ormalization +1) valueBef oreN ormalization − min


2 = max − min

valueBef oreN ormalization − min


v alueAf terN ormalization = 2 * ( max − min ) −1

x − min
or x′ = 2 * ( max − min ) − 1

3
Denormalization from range -1, +1

x = [ ( x′ 2+ 1 )(max − min) ] + min

In WEKA, for the range -1,+1, the formula is organized as follow:

x − min
x′ = 2 * ( max − min ) − 1

x − min x − min−( max−min )


x′ = ( max−min ) − 1 = [ 2
max−min ]
2 2

x − min− max min


2 + 2 ) x − max min
2 − 2 )
x′ = [ max−min ]=[ max−min ]
2 2

x − ( max min
2 + 2 )
x′ = [ max−min ]
2

x − ( max 2+ min )
x′ = max−min
2

4
Z-score standardization
Making a data set with mean=0, and standard deviation =1. This scaling method is
useful when the data follows a normal distribution (Gaussian distribution), if the data
does not follow normal distribution then this will make problems.

Example: -20, -6, 0, 40, 70,120

−20−6+0+40+70+120
M ean = 6 = 34

sd = √ (−20−34)2 +(−6−34)2 +(0−34)2 +(40−34)2 +(70−34)2 + (120−34)2


6

sd = 48.98979

z-score standardization

x−mean −20−34
x" = sd = 48.98979 = − 1.1022

Other values are changed too,

Accordingly, values are changed to:


-1.10227

-0.8165

-0.69402

0.122474

0.734847

1.755468

5
Now, if you calculate the average and sd of these new values you will see that the mean
is zero and sd=1.

Important note:

However, the point must be made that N/S are _not_ good where the raw measurement
is desirable and where the N/S is irreversible, thus losing much of the information in the
raw measurement, this is according to a note made by Kevin Hankins
([email protected]).

References
1. Yazen A. Khalil and Peshawa J. Muhammad Ali; “A proposed method for colorizing
grayscale images”, International Journal of Computer, Science and Engineering,
2013, 2(2), pp.104-109.
​http://www.iaset.us/view_acrhives.php?year=2013&id=14&jtype=2&page=2
2. Peshawa J. Muhammad Ali, Nigar M.S. Suramerry, Abdul-rahman M. Yunis, Ladeh
S.Abdulrahman, “Gender prediction of journalists from writing style”, Aro Journal,
2013, 1(1), pp.22-28. ​http://aro.koyauniversity.org/issues/volumeone/aro-10031
3. Peshawa J. Muhammad Ali; “Predicting the gender of the Kurdish writers in
Facebook” Sulaimani Journal for Engineering Sciences, 2013, 1(1), pp.18-28.
http://www.univsul.edu.iq/Wenekan_KS/12111313102014_Sulaimani%20Journal-EN
G.%2020-30.pdf

You might also like