Data

Mining
Introduction to Data Mining &
Data Preparation

Coky Fauzi Alfi

What is Data Mining?
Data mining, also known as knowledge
Knowledge
discovery from data (KDD), is the process Discovery
of extracting meaningful patterns and
information from large datasets. This
process involves using intelligent Patterns
methods to sift through vast amounts of
raw data, uncovering hidden insights,
trends, and relationships that would Large
otherwise remain obscured. Datasets
Data mining is like searching for valuable
nuggets of gold within massive amounts of
raw data. In today’s "data-rich but
information-poor" world, we’re flooded
with data, but without powerful analytical
tools, it’s challenging to extract meaningful
insights. Data mining bridges this gap by
using advanced techniques to uncover
hidden patterns, trends, and relationships
within large datasets, transforming raw
data into actionable knowledge that can
guide decisions and strategies.
Applications in Real-World Scenarios
Marketing
Businesses can use data mining to analyze customer purchase patterns, predict future buying behavior, and
tailor marketing campaigns for specific customer segments.
Healthcare
Healthcare providers leverage data mining to identify disease trends, predict patient outcomes, and develop
personalized treatment plans. Imagine analyzing medical records to identify early warning signs of a particular
illness or predicting the effectiveness of different treatments based on a patient's genetic profile and medical
history.
Finance
Financial institutions rely on data mining for fraud detection, credit risk assessment, and investment analysis.
By analyzing patterns in transactions, data mining can flag suspicious activities that may indicate fraud. It can
also help predict the likelihood of loan defaults based on a borrower's financial history and creditworthiness.
Amazon – Data Mining Process

Source: aws.amazon.com/id/data
 • IoT/Devices:
o Data from Internet of Things (IoT) devices, such as smart home gadgets, wearables, and other connected hardware.
• Applications/Logs:
o Data generated from applications, including user activity logs, clickstreams, and app usage.
• Third-Party Data:
o Data acquired from external sources or partners, which could enhance Amazon's insights.

Knowledge
Discovery

Store, Query, and Analyze

Patterns

• Database:
o Amazon Aurora and Amazon DynamoDB: These are database services for managing structured data. Aurora is a relational
database, while DynamoDB is a NoSQL database optimized for scalable, high-speed operations.
• Data Warehouse:
Large o Amazon Redshift: A data warehouse solution where large volumes of data are stored and analyzed. Redshift enables fast
querying and supports complex data analysis.
Datasets • Data Lake:
o Amazon S3: A scalable storage service often used as a data lake to store raw, structured, and unstructured data.
• Big Data:
o Amazon EMR: A tool for processing large datasets using frameworks like Hadoop and Spark, allowing for efficient analysis of big
data.
• Data Streams:
o Amazon Kinesis and Amazon MSK: Used for real-time data streaming and analytics, allowing Amazon to process live data feeds
quickly and extract instant insights.
 Data Mining vs. Machine Learning vs. AI

Aspect Data Mining Machine Learning Artificial Intelligence (AI)

Definition and Focuses on discovering hidden patterns, A subset of AI enabling computers to learn The broadest field, enabling machines to simulate
Purpose relationships, and insights from large datasets; from data without explicit programming; human intelligence, including learning, reasoning,
used for actionable knowledge extraction. builds predictive models that improve over and problem-solving.
time.

Scope Primarily an analytical process using statistical Involves creating algorithms that learn from Multiple approaches to create systems that
and machine learning techniques to understand data; narrower than AI but broader than just typically require human intelligence.
data. data mining.
Goal Extract useful information, trends, and patterns Develop models that generalize from past Create systems that mimic human intelligence
from data. data to make predictions or decisions. and perform complex tasks autonomously.

Techniques Statistical analysis, clustering, association, Algorithms such as classification, regression, Combines machine learning, neural networks,
Used anomaly detection, and machine learning clustering, and reinforcement learning. deep learning, computer vision, speech
algorithms. recognition, and rule-based systems.

Dependency May use machine learning, but also depends on Relies on data for training models and Machine learning is a primary method, but AI
statistical and analytical methods; doesn’t improved accuracy; highly data-driven. also includes rule-based systems, heuristics,
necessarily require AI. and logic.

Relationship to Often uses machine learning for data analysis A subset of AI providing tools and algorithms Broadest category, includes machine learning
Each Other but does not cover all aspects of machine frequently used in data mining. and data mining as application areas.
learning.

Examples Customer segmentation, fraud detection, Image recognition, spam filtering, personalized Autonomous driving, virtual assistants, medical
market basket analysis. recommendations. diagnosis systems.
 Data Processing Flow
Foundation Stage
At this initial stage, raw data is gathered from various sources and pre-processed. Data
Data Collection and cleaning, transformation, and integration occur here to ensure the quality and consistency
of data before analysis.
Preparation
Exploratory Stage
This stage involves analyzing prepared data to discover patterns, trends, and insights.
Techniques like clustering, association analysis, and anomaly detection help uncover
Data Mining useful information from large datasets.

Modeling and Predictive Stage

This stage takes the patterns and insights discovered in data mining to build predictive
models. Algorithms like classification, regression, and clustering learn from the data to
Machine make predictions or identify groups. This stage often includes training and testing models
Learning to ensure accuracy.

Decision-Making and Autonomous Stage

This stage leverages machine learning models, integrating them into larger systems to
enable decision-making and autonomous actions. This stage includes reinforcement
AI learning, natural language processing, computer vision, and other AI capabilities to
simulate human intelligence.
 Stages in Data Mining
#1 Data Cleaning
Removing errors, inconsistencies, and noise from the data. This
is crucial for accurate analysis, as flawed data can lead to
misleading conclusions.

Combining data from multiple sources into a unified view. This

#2 Data Integration
allows for a more comprehensive analysis, as different datasets
can provide complementary perspectives on the same
phenomenon.

Choosing the most relevant data for the specific mining task.
#3 Data Selection
This focuses the analysis on the data most likely to yield
meaningful insights.

Converting the data into a suitable format for the chosen

mining algorithms. This might involve scaling numerical data,
#4 Data
Transformation converting categorical data into numerical representations, or
creating new features based on existing ones.

Applying data mining algorithms to extract patterns and

insights from the transformed data.
#5 Data Mining
Assessing the interestingness and relevance of the discovered
patterns. This ensures that only meaningful and actionable
insights are presented to the user. Evaluation metrics like
support, confidence, and accuracy can be used to filter out
#6 Pattern trivial or unreliable patterns.
Evaluation
Assessing the interestingness and relevance of the discovered
patterns. This ensures that only meaningful and actionable
insights are presented to the user. Evaluation metrics like
support, confidence, and accuracy can be used to filter out
#7 Datasets trivial or unreliable patterns.
Overview of Data Types
Structured Data
The most common type used in data mining, structured
data is highly organized and easily searchable using
database queries. Examples include data stored in
relational databases, spreadsheets, and CSV files. This
type of data is often numerical or categorical, making it
suitable for analysis using traditional statistical and
machine learning methods.
Semi-structured Data
This type of data has some level of organization but
doesn't conform to the rigid structure of relational
databases. Examples include JSON (JavaScript Object
Notation) and XML (Extensible Markup Language)
documents, which use tags to define elements and
hierarchies within the data.
Unstructured Data
This type of data lacks a predefined structure, making it
difficult to analyze using traditional methods. Examples
include text documents, images, audio, and video.
 Example: Patient data in a healthcare system

patient_id first_name last_name age gender phone email street city state zip condition_name diagnosed_date surgery_name surgery_date allergy

123456 John Doe 45 Male -1788 [email protected] 123 Elm St Springfield IL 62701 Hypertension 6/23/2015 Appendectomy 8/20/2010 Penicillin
{
"patient_id": "123456",
"personal_info": {
"first_name": "John",
"last_name": "Doe",
"age": 45,
"gender": "Male",
"contact_info": {
"phone": "+1-555-1234",
"email": "[email protected]",
personal_info: Contains basic personal information and
"address": {
"street": "123 Elm St",
contact details, nested under the "address" key.
"city": "Springfield",
"state": "IL",
"zip": "62701"
}
}
},
"medical_history": {
"conditions": [
{
"name": "Hypertension",
"diagnosed_date": "2015-06-23"
},
{
"name": "Type 2 Diabetes",
"diagnosed_date": "2018-02-14"
} medical_history: Includes conditions, surgeries, and
],
"surgeries": [
allergies. Each entry has a name and date for context.
{
"name": "Appendectomy",
"date": "2010-08-20"
}
],
"allergies": ["Penicillin", "Peanuts"]
},
"notes": "Routine follow-up for diabetes management."
}
]
}
<patient>
<patient_id>123456</patient_id>
<personal_info>
<first_name>John</first_name>
<last_name>Doe</last_name>
<age>45</age>
<gender>Male</gender>
<contact_info>
<phone>+1-555-1234</phone> personal_info: Contains basic personal information and
<email>[email protected]</email>
<address> contact details, nested under the "address" key.
<street>123 Elm St</street>
<city>Springfield</city>
<state>IL</state>
<zip>62701</zip>
</address>
</contact_info>
</personal_info>
<medical_history>
<conditions>
<condition>
<name>Hypertension</name>
<diagnosed_date>2015-06-23</diagnosed_date>
</condition>
<condition>
<name>Type 2 Diabetes</name>
<diagnosed_date>2018-02-14</diagnosed_date> medical_history: Includes conditions, surgeries, and
</condition>
</conditions> allergies. Each entry has a name and date for context.
<surgeries>
<surgery>
<name>Appendectomy</name>
<date>2010-08-20</date>
</surgery>
</surgeries>
<allergies>
<allergy>Penicillin</allergy>
<allergy>Peanuts</allergy>
</allergies>
</medical_history>
</patient>
Data Sources
Databases
Relational databases serve as a primary source of structured data for
data mining. They store vast amounts of information in an organized
and efficient manner, making it accessible for analysis using queries
and data mining techniques.
Data Warehouses
Data warehouses consolidate data from multiple sources into a
central repository, typically structured as multidimensional data cubes
for efficient analysis and reporting. They provide a historical
perspective on data, making them valuable for trend analysis and Sources
decision-making.
Web Data
The World Wide Web has become a massive source of both structured
and unstructured data. Web pages, social media posts, online reviews,
and search queries offer valuable insights into user behavior, market
trends, and public sentiment.
Sensor Data
The sources mention sensor data as an example of streaming data,
generated continuously by devices that monitor physical phenomena.
Analyzing sensor data can reveal patterns in environmental
conditions, equipment performance, or human behavior.
 The Importance of Data Cleaning
1. Handling Missing Values
The sources describe various methods for handling missing values, which occur when there is no recorded value for a particular attribute in a data
tuple:
• Ignoring the tuple with missing values, which is ineffective unless the tuple has multiple missing values.
• Manually filling in the missing value, which is time-consuming and not feasible for large datasets.
• Using a global constant, such as "Unknown," to replace all missing values. However, this may lead the mining algorithm to mistakenly
identify the constant as an interesting concept.
• Using a measure of central tendency, such as the mean or median, to fill in the missing value.
• Using the attribute mean or median for all samples belonging to the same class as the tuple with the missing value.
• Using the most probable value, which can be determined using techniques like regression, Bayesian inference, or decision tree induction.
2. Removing Duplicates
The sources emphasize the importance of detecting and removing duplicate tuples, which can arise from data entry errors or the integration of
data from multiple sources. These inconsistencies can lead to discrepancies and affect the accuracy of the data mining results.
3. Dealing with Outliers
Outliers are data points that are significantly different from other data points in the dataset. The sources mention several techniques for
identifying and handling outliers:
• Data visualization techniques, like boxplots and scatter plots, can help identify outliers.
• Clustering, where outliers may be identified as values that fall outside of the set of clusters.
• Data smoothing techniques like binning and regression can be used to reduce the impact of outliers.
Identify Outliers Outlier

Outlier

Outlier

Outlier

Boxplot Scatter Plot

Identify Outliers
Clustering Dendrogram: Rectangle

Outlier
 Outlier
Identify Outliers
Clustering Dendrogram: Circular
Identify Outliers
Clustering Dendrogram: Phylogenic

Outlier
 Data Transformation
1. Normalization
Normalization scales attribute data to fall within a smaller range, typically [-1, 1]. This process is beneficial because the
measurement unit used for an attribute can influence the data analysis, and normalization attempts to give all attributes an
equal weight. This is particularly useful for classification algorithms involving neural networks or distance measurements, such
as nearest-neighbor classification and clustering. Three common normalization methods:
• Min-max normalization: Applies a linear transformation to map values to a new range, preserving the relationships
among the original data values.
• z-score normalization: Normalizes values based on the mean and standard deviation of the attribute, useful when the
actual minimum and maximum values are unknown or when outliers are present.
• Normalization by decimal scaling: Normalizes values by moving the decimal point, determined by the maximum
absolute value of the attribute.
2. Encoding Categorical Variables
It is a common data transformation technique used to convert categorical data (data that represents categories or groups) into
numerical representations suitable for machine learning algorithms. This information is not from the sources provided. You may
want to independently verify this information. There are various encoding methods, such as one-hot encoding and label
encoding, each with its own advantages and disadvantages.
Example: Min-max normalization
Data: [20, 35, 50, 75, 100]
𝑋 − 𝑋 𝑚𝑖𝑛
𝑋 𝑛𝑜𝑟𝑚𝑎𝑙𝑖𝑧𝑒𝑑 = × ( 𝑛𝑒𝑤 𝑚𝑎𝑥 −𝑛𝑒𝑤 𝑚𝑖𝑛 ) +𝑛𝑒𝑤 𝑚𝑖𝑛
𝑋 𝑚𝑎𝑥 − 𝑋 𝑚𝑖𝑛

𝑋 𝑚𝑖𝑛 =20 ; 𝑋 𝑚𝑎𝑥 =100 ;𝑖𝑓 𝑛𝑒𝑤𝑚𝑖𝑛 =0 ;𝑛𝑒𝑤 𝑚𝑎𝑥 =1

Data (X)
20 0
35 0.1875
50 0.375
75 0.6875
100 1

Normalized Data: [0, 0.1875, 0.375, 0.6875, 1]

Example: z-score normalization
Data: [20, 35, 50, 75, 100]
𝑋 −𝜇
𝑍= 𝑍=𝑧 𝑠𝑐𝑜𝑟𝑒𝑛𝑜𝑟𝑚𝑎𝑙𝑖𝑧𝑎𝑡𝑖𝑜𝑛;𝜇=𝑡h𝑒𝑚𝑒𝑎𝑛𝑜𝑓 𝑡h𝑒 𝑑𝑎𝑡𝑎𝑠𝑒𝑡; 𝜎=𝑡h𝑒 𝑠𝑡𝑎𝑛𝑑𝑎𝑟𝑑𝑑𝑒𝑣𝑖𝑎𝑡𝑖𝑜𝑛𝑜𝑓 𝑡h𝑒𝑑𝑎𝑡𝑎𝑠𝑒𝑡
𝜎
20+ 35+50+75 +100 280
𝜇= = =56
5 5

√ √
2 2 2 2 2
( 20 −56 ) + ( 35 −56 ) + ( 50 −56 ) + ( 75 −56 ) + ( 100− 56 ) 4070
𝜎= = =28.53
5 5
Data (X)
20 -1.26
35 -0.74
50 -0.21
75 0.67
100 1.54

Normalized Data: [-1.26, -0.74, -0.21, 0.67, 1.54]

Example: Normalization by decimal scaling
Data: [20, 350, 500, 750, 1000]
𝑋
𝑋 𝑛𝑜𝑟𝑚𝑎𝑙𝑖𝑧𝑒𝑑 = 𝑗
;

Determine 𝑗:
10

because

Data
20 0.002
350 0.035
500 0.05
750 0.075
1000 0.1

Normalized Data: [0.002, 0.035, 0.05, 0.075, 0.1]

Example: Encoding Categorical Variables
Data: ["Red", "Green", "Blue", "Green", "Red"]
If "Red" = 0; "Green" = 1; "Blue" = 2
Label Encoded Data: [0, 1, 2, 1, 0]
One-Hot Encoding creates a new binary column for each category.
Each row has a "1" in the column corresponding to its category and "0" in all other columns.
One-Hot Encoding -> "Color_Red", "Color_Green", and "Color_Blue".

Data Color_Red Color_Green Color_Blue

Red 1 0 0
Green 0 1 0
Blue 0 0 1
Green 0 1 0
Red 1 0 0
Hands-On Exercise
Install:
Thank you