Overview
Data classification is the systematic grouping of data according to shared characteristics. Different classifications answer different questions. For example, quantitative and qualitative classifications describe the nature of values, nominal and ordinal classifications describe measurement scales, and structured and unstructured classifications describe how data is organized. A dataset can belong to several classifications at the same time. Correct classification is important because analytical methods depend on what the data represents and how it was collected.
Classification By Nature
Quantitative Data
Definition
Quantitative data represents numerical quantities that can be counted or measured.
Characteristics
Values are expressed numerically and represent quantities.
Arithmetic operations may be meaningful depending on the measurement scale.
Quantitative data can be summarized using measures such as mean, median, range, variance, and standard deviation when appropriate.
It may represent counts or measurements.
Subtypes
Discrete Data
Data that takes distinct, countable values, usually whole-number counts.
Continuous Data
Data that can theoretically take any value within a specified range, depending on measurement precision.
Examples
Number of students in a class.
Number of network requests received in one minute.
Temperature measured in degrees Celsius.
Distance travelled by a vehicle.
Mass of a laboratory sample.
Qualitative Data
Definition
Qualitative data represents categories, labels, descriptions, or attributes rather than numerical quantities.
Characteristics
Values describe membership in a category or a descriptive property.
Categories may be unordered or may follow a meaningful order.
Frequency counts and category proportions are often useful summaries.
Arithmetic operations on category labels are generally not meaningful.
Subtypes
Nominal Data
Categories without an inherent ranking or natural order.
Ordinal Data
Categories that have a meaningful order, although the differences between successive categories may not be equal or measurable.
Examples
Blood group.
Species name.
Country of residence.
Product category.
Customer satisfaction level expressed as ordered categories.
Classification By Measurement Scale
Nominal Scale
Definition
A measurement scale in which observations are assigned to categories without an inherent numerical order.
Properties
Categories are mutually distinguishable according to the variable definition.
Numerical codes, if used, are labels rather than quantities.
The mode and category frequencies are meaningful summaries.
Suitable Analysis
Frequency distribution.
Proportions and percentages.
Mode.
Contingency tables.
Suitable categorical association tests.
Examples
Blood group.
Eye colour.
Country code.
Type of operating system.
Limitation
Calculating an arithmetic mean of arbitrary category codes is not meaningful.
Ordinal Scale
Definition
A measurement scale in which categories have a meaningful order but the distances between adjacent categories are not necessarily equal.
Properties
Values can be ranked.
The direction of ordering has meaning.
The numerical distance between categories may not be defined.
Rank-based summaries and methods may be appropriate.
Suitable Analysis
Frequency distribution.
Median category.
Percentiles or ranks where appropriate.
Ordinal association methods.
Ordinal regression for suitable modelling problems.
Examples
Low, medium, and high risk.
Beginner, intermediate, and advanced skill levels.
Customer satisfaction categories from very dissatisfied to very satisfied.
Disease severity categories.
Limitation
Treating ordinal categories as equally spaced numerical values requires justification and may distort interpretation.
Interval Scale
Definition
A numerical measurement scale with ordered values and meaningful, equal differences between values, but without a true zero representing the complete absence of the measured quantity.
Properties
Differences between values are meaningful.
Addition and subtraction of measurements can be meaningful.
Ratios are not generally meaningful because the zero point is arbitrary.
The mean and standard deviation can be useful summaries when their assumptions are appropriate.
Examples
Temperature measured in degrees Celsius.
Temperature measured in degrees Fahrenheit.
Calendar years under a defined calendar system.
Limitation
A value of 20 °C is not twice as hot as 10 °C because the Celsius scale does not have an absolute zero.
Ratio Scale
Definition
A numerical measurement scale with ordered values, equal intervals, and a meaningful zero representing the absence of the measured quantity.
Properties
Differences between measurements are meaningful.
Ratios between positive values are meaningful within the measurement definition.
A true zero point is present.
Many arithmetic operations are meaningful.
Examples
Mass.
Length.
Elapsed time.
Distance.
Absolute temperature measured in kelvin.
Limitation
The interpretation of a ratio still depends on the measurement definition, units, and physical context.
Classification By Structure
Structured Data
Definition
Data organized according to a predefined schema, commonly using rows, columns, fields, and data types.
Characteristics
Records follow a consistent organizational format.
Fields have defined meanings and often specified data types.
Data can be queried and validated using predefined rules.
Relational database systems commonly store structured data.
Examples
Customer tables.
Inventory records.
Examination result tables.
Financial transaction tables.
Semi Structured Data
Definition
Data that does not necessarily follow a fixed relational table structure but contains labels, keys, tags, or other organizational markers.
Characteristics
Fields may vary between records.
Hierarchical or nested structures may be present.
Metadata or tags provide organizational information.
Flexible formats support changing data structures.
Examples
JSON documents.
XML documents.
Many log formats.
Email messages with headers and body content.
Unstructured Data
Definition
Data that does not follow a predefined tabular schema and may require specialized methods for storage, search, extraction, or analysis.
Characteristics
May contain complex content without fixed rows and columns.
Can include text, images, sound, video, and documents.
Often requires specialized processing techniques.
Metadata may still be used to organize and retrieve the content.
Examples
Photographs.
Audio recordings.
Videos.
Free-form documents.
Scanned pages.
Classification By Source
Primary Data
Definition
Data collected directly by a researcher or organization for a specific investigation or purpose.
Methods
Experiments.
Surveys and questionnaires.
Interviews.
Direct observations.
Sensor measurements.
Advantages
Collection can be designed around the specific research question.
Researchers can define measurement procedures and variables.
The collection process can be documented directly.
Limitations
Collection may require substantial time and resources.
Sampling and measurement errors can still occur.
Ethical approval, participant consent, and privacy safeguards may be necessary.
Secondary Data
Definition
Data that was originally collected by another person, organization, or system and is reused for a different or additional purpose.
Sources
Published research datasets.
Government statistical repositories.
Public databases.
Historical records.
Organizational reports.
Advantages
Can reduce collection time and cost.
May provide access to large historical datasets.
Can support comparisons across regions or periods.
Limitations
The original collection method may not match the new research question.
Variable definitions and measurement procedures may be unclear.
The data may contain missing values, bias, or outdated information.
Access and reuse may be restricted by licensing, privacy, or governance rules.
Classification By Time
Cross Sectional Data
Definition
Data collected from multiple units, individuals, locations, or objects at approximately one point or period in time.
Purpose
Used to compare characteristics across different units at a particular time.
Example
Household income measurements collected from a population survey during a particular year.
Time Series Data
Definition
Data collected sequentially over time for one or more variables, usually at defined time intervals or observation times.
Purpose
Used to examine trends, seasonal patterns, temporal dependencies, and changes over time.
Example
Daily electricity consumption measurements over several years.
Longitudinal Data
Definition
Data collected repeatedly from the same subjects, units, or entities over an extended period.
Purpose
Used to investigate changes within individuals or entities and relationships between variables over time.
Example
Repeated health measurements from the same participants during a long-term study.
Panel Data
Definition
Data containing repeated observations across multiple entities and multiple time periods.
Purpose
Used to analyze both differences between entities and changes within entities over time.
Example
Annual economic indicators collected for several countries across multiple years.
Classification By Collection Method
Observational Data
Data collected by observing a system or population without deliberately assigning an experimental treatment.
Experimental Data
Data generated through an experiment in which researchers deliberately control or assign selected conditions according to a research design.
Simulated Data
Data generated by a mathematical or computational model rather than directly measured from the real-world system.
Transactional Data
Data produced through recorded activities such as purchases, payments, bookings, and service interactions.
Sensor Generated Data
Data automatically produced by instruments or devices that measure physical, environmental, biological, or operational quantities.
Classification By Granularity
Individual Level Data
Data describing individual entities, observations, or events.
Aggregated Data
Data summarized across groups, periods, locations, or other categories.
Example
Individual daily electricity readings are individual-level observations, while monthly regional electricity consumption totals are aggregated data.
Analytical Consideration
Aggregation can simplify analysis but may hide variation within groups and can create misleading interpretations if group-level results are applied directly to individuals.
Classification By Dimensionality
Univariate Data
Data involving one variable or attribute under examination.
Bivariate Data
Data involving two variables examined together to investigate relationships or differences.
Multivariate Data
Data involving three or more variables analyzed jointly to understand complex relationships and interactions.
High Dimensional Data
Data containing a very large number of variables or features relative to the number of observations, often requiring specialized statistical or computational methods.
Classification By Sensitivity
Public Data
Data that is legally and appropriately available for public access and reuse under its applicable conditions.
Internal Data
Data intended for use within an organization or defined group and not necessarily approved for public distribution.
Confidential Data
Data whose disclosure is restricted because it could harm individuals, organizations, or legitimate interests.
Personal Data
Information relating to an identified or identifiable individual, subject to applicable privacy requirements.
Sensitive Personal Data
Personal information requiring additional protection under applicable legal or organizational frameworks.
Classification Principle
Sensitivity classifications depend on the nature of the information, applicable laws, organizational policies, consent, and the potential consequences of unauthorized disclosure.
Classification By Analytical Role
Feature Variables
Variables used as inputs or predictors in an analytical model.
Target Variable
The outcome variable a supervised learning model is designed to estimate or classify.
Identifier Variables
Fields used to distinguish records or entities, such as record identifiers. They are not automatically meaningful predictive features.
Metadata
Information describing the origin, structure, meaning, collection conditions, or processing history of a dataset.
Derived Variables
Variables calculated or transformed from existing observations.
Example
In a model predicting house prices, floor area and location may be features, while the observed sale price is the target variable.
Classification And Method Selection
Principle
Data classification should guide analytical decisions. The appropriate method depends on the variable's meaning, measurement scale, collection process, research question, sample characteristics, and model assumptions.
Examples
| Data Type | Possible Methods |
|---|---|
Nominal categorical data |
|
Ordinal data |
|
Continuous numerical data |
|
Time-series data |
|
Unstructured text |
|
Image data |
|
Worked Example
Problem
A university maintains a dataset containing student identifiers, examination scores, satisfaction ratings, course names, attendance counts, and examination dates.
Classification
| Variable | Classification |
|---|---|
Student identifier | Identifier variable; represented as a categorical label rather than a numerical measurement. |
Examination score | Quantitative data. Its measurement interpretation depends on the scoring system. |
Satisfaction rating | Ordinal qualitative data when expressed through ordered categories. |
Course name | Nominal qualitative data. |
Attendance count | Discrete quantitative data. |
Examination date | Temporal data. |
Interpretation
The same dataset contains several different data types. Each variable requires an appropriate representation, summary method, and analytical treatment. Treating every column as an ordinary numerical variable would produce incorrect or meaningless calculations.
Common Classification Errors
Treating numerical category codes as actual numerical quantities.
Assuming every numerical variable is continuous.
Assuming ordinal categories always have equal intervals.
Using arithmetic averages for nominal category labels.
Ignoring the distinction between observational and experimental data.
Treating aggregated data as though it describes every individual in a group.
Assuming structured data is automatically accurate or reliable.
Ignoring data sensitivity and access restrictions during collection and processing.