DATA MINING METHODS, MODELS AND TECHNIQUES

1. Data Mining Tasks

In data mining, tasks are divided into two categories (Figure 1).
• Predictive - Make predictions about data values using known results from various
databases. Predictive data mining is based on the use of other historical data.
• Descriptive - Identifies patterns or relationships in data without any prior knowledge. It
explores the properties of the data examined. Identifies patterns or relationships in data
without any prior knowledge.

Figure 1: Data Mining Tasks

1.1 Predictive
Classification – Determines the class of an object based on its attributes using a function or a
model (Figure 2). Classification maps data into predefined groups or classes. It is often called
supervised learning because the classes are determined before examining the data. Example of
classification application to determine whether to make a bank loan and identify credit risks.
Pattern recognition is a type of classification where an input pattern is classified into one of several
classes based on similarity.
Example - An airport security screening station is used to determine whether passengers are
potential terrorists or criminals. Each passenger is scanned, and its basic pattern (distance
between eyes, size, the shape of the mouth, the shape of the head, etc.) is identified. The
database compares this pattern to find any patterns associated with known offenders.
Regression – Maps a data item to a predicted variable. Regression assumes that the target data fit
into some known type of function (e.g., linear, logistic, etc.) and determines the best function that
models the given data. Error analysis is used to determine which function is the “best”.
Example
A college professor wants to reach a certain level of savings before his retirement. Periodically, he

1
Dr. Saswati Mukherjee, School of Education Technology, JU
 wants to predict his retirement savings based on their current and several past values. He uses a
simple linear regression formula to predict the values in the future.
Time Series Analysis – The attribute’s value is examined as it varies over time. The values obtained
are evenly spaced time points (daily, weekly, hourly, etc.). Three basic functions are performed in
time series analysis. In the first case, the distance measure determines the similarity between
different time series. In the second case, the structure of the line is examined to determine (and
perhaps classify) its behaviour. A third application is to use the historical time series plot to predict
future values.
Example
The students’ examination scores were evaluated for eight consecutive academic years. The study
revealed from the preliminary analysis that marks obtained by students are below the average
mark, and further study revealed from the five-year forecast into the future that there would be
very little increase in students’ scores in the subject unless the teaching methods are modified as
per the learning habits.
Prediction –Prediction is a data mining task that is different from the prediction model, although
the prediction task is a type of prediction model. The difference is that prediction predicts a future
rather than a current state. In this case, a type of application is referred to rather than a data
mining modelling approach. Prediction applications include speech

Figure 2: Examples of data mining tasks

recognition, machine learning, and pattern recognition. Future values may be predicted using time
series analysis or regression techniques.
Example
A predictive model determines when an individual will likely complete their academic degree.
Applying this model to individual students will provide insight into when they might complete their
degrees, assuming no intervention strategy is employed. Thus, while a predictive model needs to
generate accurate scenarios, these models are not generally deployed without an intervention or

2
Dr. Saswati Mukherjee, School of Education Technology, JU
 remediation strategy in mind.

1.2 Descriptive
Clustering is similar to classification except that the groups are not predefined but rather defined by
the data alone. Clustering is alternatively referred to as unsupervised learning or segmentation. It
can be considered partitioning or segmenting the data into groups that might or might not be
disjointed. Clustering is accomplished by determining the similarity among the data on predefined
attributes.
Example
Cluster analysis was performed by organising collections of patterns into groups based on similar
student behaviour using course materials. The decision tree was the method of interest for
generating a representation of decision-making that allowed defining classes of objects for a deeper
analysis of how students learned.
Summarisation maps data into subsets with associated simple descriptions, also known as
characterisation or generalisation. It extracts or derives representative information from the
database. This may be accomplished by retrieving portions of the data or summary-type information
(the mean of some numeric attribute) from the data.
Example
Criteria, such as the average SAT or ACT scores, are used to compare universities and estimate the
type and intellectual level of the student body.
Association Rules or link analysis, also known as association or affinity analysis, is a data mining
method that uncovers patterns in a database by identifying frequently occurring relationships among
different items. These association rules are used in various fields, such as business, to aid in
marketing and decision-making by revealing patterns, e.g., customers buying shampoo also buying
conditioner. The rules are expressed as X=>Y, with X and Y representing item sets in the database.
Association rule mining involves two key measures: support (s) and confidence (c) and is conducted
in two basic steps.
• All the frequent items are found using minimum support.
• Using these frequent itemset, strong association rules are generated, having confidence c.
The first step is difficult as it demands searching all the frequently occurring itemset.
Example
A grocery store retailer is trying to decide whether to put bread on sale. To determine the impact of
this decision, the retailer generates association rules that show what other products are frequently
purchased with bread. He finds that 60% of the time, bread is sold along with pretzels, and jelly is
sold 70% of the time. Based on these facts, he tries to associate bread, pretzels, and jelly by placing
some pretzels and jelly at the end of the aisle where the bread is placed. Users of association rules
are not causal relationships. They do not represent any relationship inherent in the actual data.
Probably, no relationship between bread and pretzels causes them to be purchased together.
3
Dr. Saswati Mukherjee, School of Education Technology, JU
However, association rules can assist retail store management in advertising, marketing, and
inventory control.
Sequence Discovery - Sequential analysis or sequence discovery is used to determine sequential
patterns in data. These patterns are based on a time sequence of actions. These patterns are similar
to associations in that data (or events) are found to be related, but the relationship is based on time.
Unlike a market basket analysis, which requires the items to be purchased simultaneously, in
sequence discovery, the items are purchased over time in some order.
Example
The webmaster at XYZ Corp. periodically analyses the web log data to determine the access
frequency of XYZ’s Web pages. He is interested in determining what sequences of pages are
frequently accessed. He determines that 70 per cent of the users of page A follow one of the
following patterns of behaviour: (A, B, C) or (A, D, B, C) or ( A, E, B, C). He determines to add a link
directly from page A to page C.

2. Types of Data
The data can be divided into two categories as follows.
Categorical data are values or observations divided into nominal and ordinal. A nominal variable has
no intrinsic order that is identified in its category. An ordinal variable has a predetermined order.
Numerical data are values or observations that come from measurements. There are two types of
numerical values: discrete and continuous numbers. Discrete values are values that can be counted,
distinct, and separated from each other. Continuous values, on the other hand, are values produced
by measurements or observations that assume any value within a defined range.

3. Data Analysis Process

Data analysis is a process in which the raw data are transformed to produce visualisations and can
make predictions based on the collected data. Data analysis consists of the following stages.
a) Problem definition - Data analysis starts with a problem to be solved, which needs to be defined.
The problem definition and the corresponding documentation (deliverables) of the scientific
problem or business are very important in focusing the entire analysis strictly on getting results.
b) Data gathering/extraction - When building predictive models and conducting data analysis, it's
vital to carefully select and gather data from various sources such as log files, quizzes/tests, and
portfolios. This data includes information about students, instructors, learning materials, and
communication, as well as records of student activities. Log files are commonly used to record
interactions and are valuable for debugging. Web server log files typically contain access date
and time, request IP address, request method, and file name. Quiz and test data are organized in
a matrix form, such as the score matrix, which contains student scores for a set of questions.

4
Dr. Saswati Mukherjee, School of Education Technology, JU
 Table 1: Score/Data Matrix of Students’ Ratings

A portfolio, or e-portfolio, is a comprehensive record that provides evidence of a student's

learning progress. It includes all records of the student's activities during the learning process
and details like learning path, preferred learning styles, learning time, course grade, and
difficulty.
c) Data cleaning involves detecting and discarding erroneous or irrelevant data. The most common
inaccuracies occur due to missing data, outliers, and inconsistent/noisy data. Some possible
solutions are using a global constant to fill in the missing value or a substitute value like the
attribute mean or the mode. For example, missing values are replaced using linear interpolation
of the previous and posterior four values for emotion detection in an educational scenario or by
determining the most probable value to fill in the missing value using regression. The tuples that
contain several attributes with missing values can be removed. However, this approach is
ineffective if the percentage of missing values per attribute varies considerably. Missing values
can also be filled using the most probable value using Bayesian formalism or decision tree
induction. In educational data, missing values appear when students have not completed all the
activities in the course or when data from different sources are combined and students have
skipped some of the tasks. For example, in e-learning courses, it is often observed that some
users enrolled in a specific course once (by error or by selecting one specific activity) but never
continued the course. Noisy instances are significantly different from the remaining data set and
are called outliers. Outliers are caused by measurement errors or may result from inherent data
variability. The following are the data smoothing techniques to remove the noisy data.
• Binning methods: Binning methods smooth a sorted data value by consulting the
neighbourhood or surrounding values. The sorted values are distributed into several 'buckets'
or bins.
• Clustering: Clustering, in which similar values are organized into groups or clusters, may
detect outliers.
d) Data transformation—Derives new attributes from available attributes and facilitates a better
interpretation of information. Examples of transformation include normalisation, discretization,
derivation, and generalization.
• Normalization is a data transformation technique in which the attribute values are scaled
within a specified range, usually from -1.0 to 1.0 or between 0.0 and 1.0. In clustering, data is
normalised to avoid clusters dominated by attributes with large variations. The most

5
Dr. Saswati Mukherjee, School of Education Technology, JU
 commonly used method in education is Min-max normalisation, which performs a linear
transformation of the original data.
• Discretisation/Smoothing divides numerical data into categorical classes that are more user-
friendly than precise magnitudes and ranges. It reduces the number of possible values of the
continuous feature and provides a comprehensible view of the data. Generally, discretisation
smooths out the effect of noise and enables simpler models, which are less prone to
overfitting. Such techniques include binning, clustering, and regression. A special type of
discretisation is the transformation of ordinal to binary representation.
e) Derivation is a technique used to transform data by creating new attributes from existing ones.
This involves creating a new attribute based on one or more existing attributes in the dataset.
The new attribute could be derived through a mathematical transformation of another attribute,
such as converting a time difference attribute from seconds to minutes. Another commonly used
type of derivation involves performing aggregation on another attribute. For example, a hash
code can be used to combine student information into a single hash number by multiplying each
field with a distinct power of ten in descending order. Some examples of derived attributes from
e-learning system data are shown in Table 2.
In data generalization, raw data is replaced by higher-level concepts using concept hierarchies.
For instance, categorical attributes like "street" can be generalized to higher-level concepts like
"city" or "country". Similarly, numeric attributes like "age" can be mapped to higher-level
concepts such as "young", "middle-aged", and "senior".

Table 2: Example of Derived Attributes

Attribute Description
UserId A unique identifier per user
Performance Percentage of correctly answered tests calculated as the
number
of correct tests
TimeReading Time spent on pages (calculated as the total time spent on
each
page accessed) in
NoPages The number of accessed pages
TimeTests The time spent performing tests (calculated as the total time
spent
on each test)
Motivation Engaged/Disengaged

f) Data integration involves combining data from various sources into a consistent data store, such
as a data warehouse. These sources can include multiple databases, data cubes, or flat files.
There are several important issues to consider during data integration. One major issue is entity
identification, where the data analyst must verify that the customer ID in one database and the
6
Dr. Saswati Mukherjee, School of Education Technology, JU
 cust_number in another refer to the same entity. Databases and data warehouses typically
contain metadata, which is data about the data. This metadata can be used to prevent errors in
schema integration. Redundancy is another important issue, where an attribute is derived from
another table. Inconsistencies in attribute or dimension naming can also lead to redundancies in
the resulting data set.
g) Data reduction techniques help in analysing the reduced representation of the dataset without
compromising the integrity of the original data and yet producing quality knowledge. The
concept of data reduction is commonly understood as either reducing the volume or reducing
the dimensions (number of attributes). Several methods have facilitated the analysis of a
reduced volume or dimension of data and yet yield useful knowledge. Strategies for data
reduction include the following.
• Data cube aggregation is applied to the data when constructing a data cube.
• Dimensionality reduction corresponds to detecting and removing irrelevant, weakly relevant,
or redundant attributes or dimensions.
• In data compression, encoding mechanisms reduce the size of the data set. The methods
used for data compression are wavelet transform and Principal Component Analysis.
• In numerosity reduction, data are replaced or estimated by smaller data representations,
such as parametric models (which need to store only the model parameters instead of the
actual data. e.g., regression and log-linear models), or nonparametric methods such as
clustering, sampling, and histograms.
h) Discretisation and concept hierarchy generation replace the raw data values for attributes with
ranges or higher conceptual levels. Concept hierarchies allow data mining at multiple levels of
abstraction.
i) Predictive modelling is a process of creating or choosing a suitable statistical model to predict
the probability of a result. These models are useful for understanding the system and are used
for two main purposes. The first is to make predictions about the data values produced by the
system using regression models. The second is to classify new data products using classification
or clustering models. The models can be divided according to the type of result that they
produce.
• Classification models: The result obtained by the model type is categorical.
• Regression models: The result obtained by the model type is numeric.
• Clustering models: The result obtained by the model type is descriptive.
j) Model validation assesses the validity of the data produced by the model by comparing them
with the actual system. Model validation quantifies the performance expected from a machine
learning model on unseen data. For this reason, model validation is often done on datasets the
model does not use during training. There are different approaches to designate data sets for
machine learning model development, such as train/validate/test with percentage splits like
(80/10/10), k-fold cross-validation, and time-based splits.

7
Dr. Saswati Mukherjee, School of Education Technology, JU
k) Deployment consists of implementing the results obtained from the data analysis. This is
translated into design solutions or scientific publications in technical or scientific environments.

Quantitative and Qualitative Data Analysis

In quantitative analysis (Figure 3), the data are logically ordered and categorized in a way that
creates structures within the data. This order, categorization, and structure provide additional
information and allow for further data processing using mathematical models to make quantitative
predictions. This enables data analysts to draw objective conclusions.
In qualitative analysis, the data lack a structure and are neither numeric nor categorical. For
example, qualitative study data could include written textual, visual, or audio data. This type of
analysis relies on methodologies, often ad hoc, to extract information that provides qualitative
predictions or subjective interpretations. Qualitative analysis explores more complex systems and
draws conclusions that are not achievable through a strictly mathematical approach.

Figure 3: Quantitative and qualitative analyses

4. Issues in Data Mining

There are several important implementation issues associated with data mining, which are outlined
below.
Overfitting: When a model is created using a particular dataset, it should be able to perform well
with new unseen data. Overfitting happens when the model does not generalize well to new data.
This can be caused by using a small training dataset. For example, consider two students, Y and Z,
preparing for an exam. Y memorizes the entire textbook while Z studies and practices all the
questions. Student Y will only be able to answer questions similar to those in the book, while student
Z will be able to answer all the exam questions. Student Y may perform well on the studied material
but poorly on new or unfamiliar questions. A model is considered a good fit when it performs well
with both the training dataset and new, unseen data, similar to student Z.
• Outliers: Many data entries often do not fit nicely in the derived model. This becomes even more
of an issue with very large databases. If a model is developed with outliers, it may not behave
well for data without outliers.
8
Dr. Saswati Mukherjee, School of Education Technology, JU
• Interpretation of results: Data mining output may require experts to interpret the results
correctly, which might otherwise be meaningless to the average database user.
• Visualisation of results: Visualising the results helps one easily view and understand the output
of data mining algorithms.
• Large datasets: The massive datasets associated with data mining create problems when
applying algorithms designed for small datasets. Many modelling applications grow exponentially
with the dataset size and are thus too inefficient for large datasets.
• High Dimensionality: Conventional database schemas can consist of numerous attributes.
However, not all attributes may be necessary for solving a data mining problem, and some may
even hinder the correct completion of the task. In addition, using too many attributes can make
algorithms more complex and less efficient. This issue is often called the "dimensionality curse,"
as it can be challenging to determine which attributes are essential. One approach to addressing
this high-dimensionality problem is to reduce the number of attributes, known as dimensionality
reduction.
• Multimedia data: Most previous data mining algorithms target traditional data types (numeric,
character, text, etc.). Using multimedia data, such as that found in GIS databases, complicates or
invalidates many proposed algorithms.
• Missing data: During the pre-processing phase of KDD, missing data may be replaced with
estimates. Handling missing data with various approaches can lead to invalid results in the data
mining step.
• Irrelevant data: Some attributes in the database might not interest the data mining task being
developed.
• Noisy data: Some attribute values might be invalid or incorrect. These values are often
corrected before running data mining applications.
• Changing data: Databases cannot be assumed to be static. However, most data mining
algorithms do assume a static database. This requires that the algorithm can be executed
anytime the database changes.
• Integration: Knowledge Discovery in Databases (KDD) is not currently integrated into regular
data processing activities. KDD requests are often seen as special, unusual, or one-time needs,
which makes them inefficient, ineffective, and not widely applicable for ongoing use. It is
desirable to integrate data mining functions into traditional DBMS systems.
• Application: Determining the intended use of the information obtained from the data mining
function is challenging. Because the data types are not always previously known, business
practices may have to be modified to determine how to use the information effectively.

5. Data Mining Methods

The techniques utilised in Educational Data Mining (EDM) and Learning Analytics (LA) are usually
drawn from data mining, data analytics, psychometrics, and educational measurement. In many
9
Dr. Saswati Mukherjee, School of Education Technology, JU
instances, the unique qualities of educational data have led to certain techniques playing a more
significant role in EDM/LA than data mining.
EDM methods may be categorised, as shown in Figure 4, based on the classifications by various
authors. However, this section describes four major classes of methods used in EDM.
• Prediction
• Structure Discovery
• Relationship Mining
• Discovery with Models

Figure 4: Categorization of EDM methods

Prediction Method
In educational data mining, prediction involves creating models to forecast various aspects of student
data using related information. The aim is to predict outcomes like potential dropouts or failures,
enabling timely intervention. There are three main types of prediction methods in educational data
mining:
1. Classifier: This predicts binary or categorical outcomes using techniques such as decision trees,
random forests, logistic regression, support vector machines, and various neural network variants.
2. Regressor: This is used to predict continuous variables, such as numbers. Linear regression is
commonly utilized in educational data mining.
3. Latent Knowledge Estimation: This method assesses students' comprehension of specific skills and
concepts by analysing their accuracy patterns in problem-solving exercises. Techniques used for latent
knowledge estimation include Naïve Bayes, Bayes net, Bayesian Knowledge Tracing (BKT), and
performance factors assessment.

10
Dr. Saswati Mukherjee, School of Education Technology, JU
Structure discovery
Structure discovery methods use an unsupervised classification mechanism to find structure in the
dataset. The data items in the dataset are classified based on the values of attributes rather than
class-label usage, as in prediction methods. Structure discovery algorithms attempt to find structure
in the data without any ground truth or a priori idea of what should be found. Common structure
discovery algorithms in educational data include the following.
• Clustering
• Factor analysis
• Domain structure discovery
• Network analysis

Clustering
Clustering methods group data items with similar attributes, dividing the dataset into clusters.
Distance measures such as Manhattan and Euclidean are employed to measure the similarity or
dissimilarity of the items. In educational data mining (EDM), clustering is used to group students with
similar intelligence and behaviour, as well as to group similar learning materials for a subject.
Clustering is especially useful when the common categories in the dataset are not known in advance.
It is commonly used to group students based on their actions, behaviors, similarities, and
differences. Additionally, clustering has been applied to study school principals' leadership
strategies, differences between schools' academic offerings, and variations in how students
approach gameplay in educational games. This approach is known as Educational Data Clustering
(EDC). In addition to grouping, EDC is employed to analyse learning styles and enhance student
collaborative learning.
Factor Analysis
Factor analysis is used to uncover the latent structure of a set of variables. It reduces attribute space
from a large number of variables to fewer factors. In educational data mining, factor analysis is used
for dimensionality reduction (e.g., reducing the number of variables) during pre-processing to reduce
over-fitting and to determine meta-features. Factor analysis helps to understand how different
students behave in Learning Management Systems by extracting factors/hidden variables from the
dataset and assessing the covariance of the factors/hidden variables. The most common approaches
for factor extraction are Principal Components Analysis, Common Factor Analysis, and Maximum
Likelihood (ML).
Domain Structure Discovery
Domain structure discovery consists of finding which items map to specific student skills. This
includes mapping problems in educational software to specific knowledge components required in a
subject area. Domain structure discovery methods map a dataset's attributes to the domain's
knowledge states. In EDM, domain structure discovery is used to evaluate the expertise in skills by
mapping key educational concepts (knowledge states) to the structure of tasks performed by
11
Dr. Saswati Mukherjee, School of Education Technology, JU
students. For example, a questionnaire on basic arithmetic with four questions can be used to map
the marks obtained by the student for the answers to the questions to specific educational concepts,
such as addition, subtraction, and multiplication.
Network Analysis
In network analysis, models are developed to find the relationships and interactions between
individual actors or elements and the patterns that emerge from those relationships and
interactions. The most common use of network analysis is in social network analysis (SNA), which
focuses on social relationships and linguistic interaction patterns.

Figure 8: Different characteristics of social networks. A, B, and C show varying centrality

and density of networks; panel D shows network closure, i.e., when two actors, tied to a
common third actor, tend to also form a direct tie between them. Panel E represents two
actors with different attributes (e.g., organizational affiliation, beliefs, gender, who tend to
form ties. Panel F consists of two types of ties: friendship (solid line) and dislike (dashed
line)

Some Definitions of Social Network

Actor is known known as node or vertex represents an object or entity that has relationships with
other object/entity.
Tie known as relation or edge, describes a particular, well specified, relationship between two
Actors. This could refer to relationships like “went to the same school” or “likes potato chips”. Ties
can be un-directed (e.g., went to the same school). Ties can also be directed either unidirectionally
or bidirectionally.
Network, also called a Graph, refers to a collection of Actors and the Ties between them.
Multiplex networks are networks where more than one kind of tie is present. For example, when
several different kinds of relationships exist between bank managers (e.g., going for advice, being
friends with, working for, etc.), then the network contains multiple tie types between actors.
Weighted Ties contain multiple different kinds of edges between actors with varying strengths. For
example, A might like B in total, but B and C only like each other moderately.
Group in a network is just a subset of the actors which share some characteristic in common. In an
organizational network, one group consists of all actors that work in the human resource
12
Dr. Saswati Mukherjee, School of Education Technology, JU
department.

Social Network Data

A Sociomatrix (also known as an Adjacency Matrix) represents directed or undirected ties between
actors using a numerical matrix. There is one column for each actor and one row for each actor. In
general, the diagonal elements of this matrix are always equal, signaling that actors do not tie
themselves (Figure 9). To specify an entry in the matrix, use the same convention: [row i, column j]
[3,5] Each row in the sociomatrix represents the ties that Actor I have with all other actors (j’s).

Figure 9: Adjacency Matrix

Figure 10: Edge List

Edgelist (Figure 10) is the other primary form of data storage for social network analysis. It captures
information about existing ties in the network. This form of data entry is best for storing information
about data collected by hand, as it is very efficient and relatively easy to enter.

Properties of Nodes
Centrality is a measure that quantifies an actor's influence in the network.
• Degree Centrality captures the number of ties to a given actor. For undirected ties, this
counts the number of ties for every actor. For directed networks, actors can have both
indegree and outdegree centrality scores. As the name implies, centrality measures how
central or well-connected an actor is in a network. This denotes power and increased access
to information. A high degree of centrality is generally considered an asset to an actor.

Figure 11: Centrality Measures

13
Dr. Saswati Mukherjee, School of Education Technology, JU
 • Betweenness Centrality is defined as the number of shortest paths between nodes that go
through a particular actor. More precisely, it is the sum of the shortest path lengths between
every set of alters where the path goes through the actor; we are calculating the measure
divided by the shortest path lengths (not necessarily through the target actor) between those
actors.
• Closeness centrality measures how many steps (ties) are required for a particular actor to
access every other actor in the network. This is measured as 1 divided by the sum of geodesic
distances from an actor to all alters in the network. The measure will reach its maximum for a
given network size when an actor is directly connected to all others in the network and its
minimum when an actor is not connected to others.
• Eigenvector centrality measures how connected an actor is to other well-connected actors
within a network, using a mathematical property of networks represented as an adjacency
matrix. This calculation determines an actor's level of connectivity to other well-connected
actors. In essence, it reflects the value of having many influential connections.
• Meanwhile, brokerage describes the advantageous position of actors who can facilitate
interactions between other actors in the network. Brokerage Centrality measures an actor's
degree of occupying such a brokerage position across all pairs of actors. It aims to capture the
idea that a broker acts as an intermediary and can benefit from their position as a go-
between.
There is no fixed methodology for Social Network Analysis. However, most approaches contain
most of the steps described below.
• STEP 1: Identify the target network and its purpose. Some work may need to be done to
define the network or to draw boundaries between where one network ends and another
one starts.
• STEP 2: Design the methodology for collection and analysis. Data for SNA is often generated
through questionnaires or surveys administered to different network actors. The type of
questions contained within the questionnaires or surveys significantly impacts the
information generated. Other potential methods for data collection include observation,
semi-structured interviews, and focus group discussions.
• STEP 3: Collect, store, and process the data using a common software package such as
Microsoft Excel or Access. In other cases, data might need to be input directly into the
specialist SNA software package.
• STEP 4: Present the data in different ways, including matrices, graphs, tables and plots.
• STEP 5: Analyze the data and recommend appropriate action

Relationship mining
Relationship mining aims to uncover connections between variables in a data set with multiple
variables. It can identify which variables are strongly linked to a specific variable of interest or
14
Dr. Saswati Mukherjee, School of Education Technology, JU
discover the strongest relationships between any two variables. In the context of Educational Data
Mining (EDM), four types of relationship mining are commonly used:
• Association rule mining originates from the field of data mining and is used for "market
basket" analysis in business data. It aims to find if-then rules, such as if a set of variable
values is found, then another variable will generally have a specific value. This type of mining
has various applications in EDM, such as representing patterns of successful student
performance.
• Sequential pattern mining investigates temporal associations between events. It is a special
case of association rule mining and is used to capture complex patterns, such as studying how
students' behaviours change over time in an inquiry learning environment.
• Correlation mining, which determines positive or negative linear correlations between
variables. Its goal is to use post-hoc analysis and/or dimensionality reduction techniques to
avoid spurious relationships. Correlation mining is used to study the relationship between
student attitudes and help-seeking behaviours.
• Causal data mining, which aims to find whether one event causes another, such as predicting
factors that may lead a student to perform poorly in a class. These methodologies have the
potential to uncover unexpected but meaningful relationships between variables and can be
used for generating new hypotheses or identifying contexts for potential intervention by
automated systems.

Discovery with models

Discovery with the model’s technique involves two phases of data analysis. In the first phase, a
model is built using either educational data mining methods or human reasoning. EDM methods like
prediction, relationship mining, and clustering can be used to build the model. The model obtained
in the first phase is used as a component/model in the second phase of complex attribute analysis.

15
Dr. Saswati Mukherjee, School of Education Technology, JU