Lecture #2

Data Categorization

Dr. Debasis Samanta

Professor
Department of Computer Science & Engineering

@DSamanta, IIT Kharagpur Data Analytics (CS61061) 1

 Quote of the day..

 The simple things are also the most extraordinary

things, and only the wise can see them.
 Be minute to everything around you. The world is a great
teacher!
 PAULO COELHO Brazillian author.

@DSamanta, IIT Kharagpur Data Analytics (CS61061) 2

We are going to learn…
 Data in data analytics

 NOIR topology

 Nominal scale of measurement

 Ordinal scale of measurent

 Interval scale of measurement

 Ratio scale of measurement

 Data model for high-dimensional data

@DSamanta, IIT Kharagpur Data Analytics (CS61061) 3

Data in Data Analytics

 Entity: A particular thing is called entity or object.

 Attribute. An attribute is a measurable or observable property of an entity.
 Data. A measurement of an attribute is called data.
 Note
 Data defines an entity.
 Computer can manage all type of data (e.g., text, numeric, image, audio,
video, etc.).
@DSamanta, IIT Kharagpur Data Analytics (CS61061) 4
Data representation
 How a document (e.g., text) can be represented?

@DSamanta, IIT Kharagpur Data Analytics (CS61061) 5

Data representation
 How an image can be represented?

@DSamanta, IIT Kharagpur Data Analytics (CS61061) 6

Data representation
 How a video can be represented?

@DSamanta, IIT Kharagpur Data Analytics (CS61061) 7

 Data representation
 How the streaming data from an artificial earth satellite can be
represented?

@DSamanta, IIT Kharagpur Data Analytics (CS61061) 8

Data in Data Analytics
 In general, there are many types of data that can be used to
measure the properties of an entity.

 A good understanding of data scales (also called scales of

measurement) is important.

 Depending the scales of measurement, different techniques

are followed to derive hitherto unknown knowledge in the
form of
 patterns, associations, anomalies or similarities from a volume of
data.

@DSamanta, IIT Kharagpur Data Analytics (CS61061) 9

 NOIR
Classification of scales of Measurement

@DSamanta, IIT Kharagpur Data Analytics (CS61061) 10

NOIR classification
 The mostly recommended scales of measurement are
N: Nominal
O: Ordinal
I: Interval
R: Ratio

The NOIR scale is the fundamental building block on

which the extended data types are built.

@DSamanta, IIT Kharagpur Data Analytics (CS61061) 11

 NOIR Classification

Nominal Ordinal Interval Ratio

Alphabetical
Binary Ternary Others
Ordered Discrete

Numerically
Symmetric Continuous
Ordered

Literally
Asymmetric
Ordered

Categorical (Qualitative) Numeric (Quantitative)

@DSamanta, IIT Kharagpur Data Analytics (CS61061) 12
Nominal scale
 Definition
A variable that takes a value among a set of mutually exclusive codes that have no logical
order is known as a nominal variable.

 Examples
Gender Used letters or numbers
{ M, F} or { 1, 0 }

Blood groups Used string

{A , B , AB , O }

Rhesus (Rh) factors Used symbols

{+ , - }

Country code ??
????

@DSamanta, IIT Kharagpur Data Analytics (CS61061) 13

Nominal scale: Properties
Note
 The nominal scale is used to label data categorization
using a consistent naming convention.
 The labels can be numbers, letters, strings,
enumerated constants or other keyboard symbols.
 Nominal data thus makes “category” of a set of data.

 The number of categories should be two (binary) or

more (ternary, etc.), but countably finite.

@DSamanta, IIT Kharagpur Data Analytics (CS61061) 14

Nominal scale: Properties
Note
 A nominal data may be numerical in form, but the numerical values
have no mathematical interpretation.
 For example, 10 prisoners are 100, 101, … 110, but; 100 + 110 = 210 is
meaningless. They are simply labels.

 Two labels may be identical ( = ) or dissimilar ( ≠ ).

 These labels do not have any ordering among themselves.
 For example, we cannot say blood group B is better or worse than
group A.

 Labels (from two different attributes) can be combined to give

another nominal variable.
 For example, blood group with Rh factor ( A+ , A- , AB+, etc.)

@DSamanta, IIT Kharagpur Data Analytics (CS61061) 15

Binary scale of nominal data
 Definition
A nominal variable with exactly two mutually exclusive categories that
have no logical order is known as binary variable

 Examples
Switch: {ON, OFF}
Attendance: {True, False}
Entry: {Yes, No}
etc.

Note
 A Binary variable is a special case of a nominal variable that
takes only two possible values.

@DSamanta, IIT Kharagpur Data Analytics (CS61061) 16

Symmetric and Asymmetric Binary Scale
 Different binary variables may have unequal importance.

 If two choices of a binary variable have equal importance,

then it is called symmetric binary variable.
 Example: Gender = {male, female}
// usually of equal probability.

 If the two choices of a binary variable have unequal

importance, it is called asymmetric binary variable.
 Example: Food preference = {V, NV}

@DSamanta, IIT Kharagpur Data Analytics (CS61061) 17

Operations on Nominal variables
 Summary statistics applicable to nominal data is mode.
 Arithmetic ( + , - , * a n d / ) and logical operations ( < , > , ≠ ,
e t c . ) are not permitted.
 The allowed operations are : accessing (read, check, etc.) and re-
coding (into another non-overlapping symbol set, that is, one-
to-one mapping), etc.
 Nominal data can be visualized using line charts, bar charts or
pie charts, etc.
 Two or more nominal variables can be combined to generate
other nominal variable.
 Example: Gender (M,F) × Marital status (S, M, D, W)

@DSamanta, IIT Kharagpur Data Analytics (CS61061) 18

Ordinal scale
 Definition
Ordered nominal data are known as ordinal data and the
variable that generates it is called ordinal variable.
 Example:
Shirt size = { S, M, L, XL, XXL}

Note
The values assumed by an ordinal variable can be ordered
among themselves as each pair of values can be compared
literally or using relational operators ( < , ≤ , > , ≥ ).

@DSamanta, IIT Kharagpur Data Analytics (CS61061) 19

Operation on Ordinal data
 Usually relational operators can be used on ordinal data.
 Summary measures mode and median can be used on ordinal data.
 Ordinal data can be ranked (numerically, alphabetically, etc.) Hence,
we can find any of the percentiles measures of ordinal data.
 Calculations based on order are permitted (such as count, min, max,
etc.).
 Spearman’s R can be used as a measure of the strength of association
between two sets of ordinal data.
 Numerical variable can be transformed into ordinal variable, but with a
loss of information.
 For example, Age [1, … 100] = [young, middle-aged, old]

@DSamanta, IIT Kharagpur Data Analytics (CS61061) 20

Interval scale
 Definition
It allows to measure the interval between two measures.

Interval scale data are like ordinal data, in that they can be placed in a
meaningful order. In addition, they have meaningful intervals between them.

Example 1:
S, M, L, being in ordinal scale, we cannot say that interval between S and M is
same as that of between M and L , etc. Whereas, on the Celsius scale (which
is an interval scale of measurement), the difference between 100oC and 90oC
is the same as the difference between 50oC and 40oC.

Note that in interval scale of measurement, a zero-value does not mean that
there is nothing!

@DSamanta, IIT Kharagpur Data Analytics (CS61061) 21

Interval scale: Properties
 Examples
Latitude, longitude, temperature (in Celsius/Fahrenheit scale),
calendar dates, etc.

Properties
 Interval data are with well-defined interval.
 Interval data are measured on a numeric scale (with +ve, 0
(zero), and –ve values).
 Interval data may have a zero point on origin. However, the
origin does not imply a true absence of the measured
characteristics.
 For example, the temperature outside is 0oC. Here, 0oC does not
indicate a complete absence of heat; it is a value of a temperature.

@DSamanta, IIT Kharagpur Data Analytics (CS61061) 22

Operations on Interval data
 We can add to or from interval data.
 For example: date1 + x-days = date2
 Subtraction can also be performed.
 For example: current date – date of birth = age
 Negation (changing the sign) and multiplication by a
constant are permitted.
 All operations on ordinal data defined are also valid here.
 Linear (e.g. cx + d ) or Affine transformations are
permissible.
 Other one-to-one non-linear transformation (e.g., log, exp,
sin, etc.) can also be applied.
@DSamanta, IIT Kharagpur Data Analytics (CS61061) 23
Operation on Interval data
Note
 Interval data can be transformed to nominal or ordinal
scale, but with a loss of information.

 Interval data can be graphed using histogram,

frequency polygon, etc.

 The statistical estimation like mean, median, and

mode can be calculated.

@DSamanta, IIT Kharagpur Data Analytics (CS61061) 24

Ratio scale
 Definition
Interval data with a clear definition of “zero” are called ratio data.
 Examples:
Temperature in Kelvin scale, intensity of earth-quake on Richter scale, sound intensity
in Decibel, cost of an article in Rupees, population of a country, weight of a body, age of
a tree, height of a building, etc.

Note
 The data with ratio scale of measurement are the mostly used data in data
science.
 In ratio scale, both differences between data values and ratios (of non-zero)
data pairs are meaningful.
 100oC is not twice as hot as 50oC. On the other hand, 100 Kg is twice heavy as
50 Kg.
 Temperature in Kelvin scale is a ratio scale of measurement. Here, 0oK means
an absolute temperature and also we can say that 20oK is as twice as 10oK.

@DSamanta, IIT Kharagpur Data Analytics (CS61061) 25

Ratio scale: Properties
 Properties

 All ratio data are interval data but the reverse is not true.
 Ratio scale, as mentioned earlier has an absolute zero
characteristic. It has orders and equally distanced value between
units. The zero point characteristic makes it relevant or
meaningful to say, “one object has twice the length of the other”
or “is twice as long”.
 Ratio scale doesn’t have a negative number, unlike interval scale
because of the absolute zero or zero point characteristic.

To measure any object on a this scale, researchers must first see if

the object meets all the criteria for interval scale plus has an
absolute zero characteristic.

@DSamanta, IIT Kharagpur Data Analytics (CS61061) 26

Operation on Ratio data
 All arithmetic operations on interval data are
applicable to ratio data.

 In addition, multiplication, division, etc. are allowed.

 Mean, median and mode are the permissible statistical

operations.

 Any linear transformation of the form ( ax + b )/c are

known.

@DSamanta, IIT Kharagpur Data Analytics (CS61061) 27

Properties of data
 Following FOUR properties (operations) of data are pertinent.

# Property Operation Type

1. Distinctiveness = and ≠
Categorical
(Qualitative)
2. Order <,≤,>,≥

3. Addition + and -
Numerical
(Quantitative)
4. Multiplication * and /
@DSamanta, IIT Kharagpur Data Analytics (CS61061) 28
NOIR summary
 Nominal (with distinctiveness property only)

 Ordinal (with distinctive and order property only)

 Interval (with additive property + property of Ordinal data)

 Ratio (with multiplicative property + property of Interval

data)

 Further, nominal and ordinal are collectively referred to as

categorical or qualitative data. Whereas, interval and ratio
data are collectively referred to as quantitative or numeric
data.

@DSamanta, IIT Kharagpur Data Analytics (CS61061) 29

 Data Cube
Multidimensional Data Modeling

@DSamanta, IIT Kharagpur Data Analytics (CS61061) 30

Concept of data cube
 A multidimensional data model views data in the form
of a cube.

 A data cube is characterized with two things

 Dimension: the perspective or entities with respect to
which an organization wants to keep record.
 Fact: The actual values in the record

Example.
 Rainfall data of Metrological Department
 Time (Year, Season, Month, Week, Day, etc.)
 Location (Country, Region, State, etc.)
@DSamanta, IIT Kharagpur Data Analytics (CS61061) 31
2-D view of rainfall data

 In this 2-D representation, the rainfall for “North-East”

region are shown with respect to different months for a
period of years

@DSamanta, IIT Kharagpur Data Analytics (CS61061) 32

3-D view of rainfall data
 Suppose, we want to represent data according to times
(Year, Month) as well as regions of a country say East,
West, North, North-East, etc.

 A 2-D view of 3-D rainfall data

@DSamanta, IIT Kharagpur Data Analytics (CS61061) 33

3-D view of rainfall data

 Data cube: This enables us a 3-D view of the rainfall data

@DSamanta, IIT Kharagpur Data Analytics (CS61061) 34

3-D view of rainfall data

India China Russia Pakistan

 Data cube: This enables us a 3-D view of the rainfall data

for a continent say?
@DSamanta, IIT Kharagpur Data Analytics (CS61061) 35
3-D view of rainfall data

 What is the data cube representation of rainfall data of

the entire world?

@DSamanta, IIT Kharagpur Data Analytics (CS61061) 36

Data cube aggregation
ROLL UP

DRILL DOWN
@DSamanta, IIT Kharagpur Data Analytics (CS61061) 37 37
Data cube segregation

BASE CUBOID

SLICE
@DSamanta, IIT Kharagpur Data Analytics (CS61061) 38 38
 Reference

 The detail material related to this lecture can be found in

Data Mining: Concepts and Techniques (3rd Edn.) by Jiawei Han,

Michelline Kamber and Jian Pei, Morgan Kaufmann (2014).

@DSamanta, IIT Kharagpur Data Analytics (CS61061) 39

 Any question?

@DSamanta, IIT Kharagpur Data Analytics (CS61061) 40

 Questions of the day…
1. Consider an image as an entity.
• What are the attributes you should think to represent an
image?
• Categorize each attribute according to the NOIR data
classification.
• Suppose, two images are given. Give an idea to check if two
images are identical or not.

2. How you can convert a data of interval type to ordinal

type? Give an example. What are the issues of such
transformation? Whether the reverse is possible or not?
Justify you answer.

@DSamanta, IIT Kharagpur Data Analytics (CS61061) 41

 Questions of the day…
3. What are the different properties used to categorize
the data according to NOIR data categorization?

4. Given an entity say “STUDENT” with the following

attributes. Identify the NOIR category to which each
of them belongs.
Scholarship Name RollNo DoB Aaadhar Gender Mobiloe Email Id
amount No. No.

@DSamanta, IIT Kharagpur Data Analytics (CS61061) 42

 Questions of the day…
5. Give the concept of data cube to represent hyper-
dimensional data? Also, explain with suitable
diagrams the following.
 Roll up
 Drill down
 Slice
6. Using the concept of data cube, how YouTube can
archive videos of all type?
7. Give FOUR differences between data of types
“interval” and “ratio-scale”

@DSamanta, IIT Kharagpur Data Analytics (CS61061) 43

 Questions of the day…
8. What are the different types of data you can think to
judiciously represent an entity like the following?

@DSamanta, IIT Kharagpur Data Analytics (CS61061) 44