BIT students’ performance analysis with KNIME Analytics

Platform

Enxhi Nikolla

Bachelor’s Thesis
Degree Programme in Busi-
ness Information Technology
 Abstract

30.11.2018

Author(s)
Enxhi Nikolla
Degree programme
Business Information Technology
Report/thesis title Number of pages
and appendix pages
BIT students’ performance analysis with KNIME Analytics Platform 48 + 1

The ability to analyse students’ behaviours and make discussions, recommendation or fu-
ture predictions is the core idea of Educational Data Mining. Universities are more and
more interested to implement this data mining technique for better students’ performance
and good school reputation.

In this thesis, we are analysing students’ behaviours of Business Information Technology

degree program in Haaga-Helia University of Applied Sciences to assess current trends for
future recommendations and improvements.

There is provided valuable information for understanding the environment we are working
on such as statistical analysis, machine learning, data mining and educational data mining.

Two iterations will be implemented for reaching the desired level of results in this analysis.
The first iteration will provide a brief description on the KNIME Analytics Platform project
development with the data collected from the questionnaire. On the second iteration, the
results provided by the first iteration will be interpreted into valuable conclusions, observa-
tions as well as suggestions on student’s performance. Interpretation is done based the
Niemivirta study of 8 scale factors of student performance.

This thesis is target to Haaga-Helia University of Applied Sciences’ pedagogical and ad-
ministrative staff, academic advisors and students. However, psychologists, data analysers
and anyone else who is interested in Educational Data Mining and students’ behaviours
can find this thesis useful.

To summarize the value of this thesis is to give a big picture of the current level of student
performance for the given dataset and analyse how this performance is affected from each
scale factor mentioned in the Niemivirta study.

Keywords
Machine Learning, Data Mining, KNIME, Education, EDM
Table of Contents

Acknowledgements............................................................................................................ 4
Abbreviations ..................................................................................................................... 5
1 Introduction ................................................................................................................... 1
2 Research Question ....................................................................................................... 2
2.1 Sub-Questions ...................................................................................................... 2
3 Methods ........................................................................................................................ 3
3.1 Methodology ......................................................................................................... 3
3.2 Mixed Methods:..................................................................................................... 3
3.3 Workflow Processes.............................................................................................. 3
4 Background Studies ...................................................................................................... 7
4.1 Statistical Analysis ................................................................................................ 7
Statistical analysis methods .................................................................................. 8
Statistical analysis techniques............................................................................. 10
4.2 Data Mining ......................................................................................................... 13
Data mining in years and now ............................................................................. 13
Importance of Data Mining .................................................................................. 14
Data Mining Methods .......................................................................................... 14
How does Data Mining work?.............................................................................. 15
4.3 Machine Learning ............................................................................................... 15
History of Machine Learning................................................................................ 16
What is Machine Learning? ................................................................................. 18
Types of Machine Learning ................................................................................. 19
Automatic Tools for Machine Learning and Data Mining ..................................... 21
Machine Learning and Data Mining main tools in years ...................................... 21
4.4 Deep Learning .................................................................................................... 27
Educational Data Mining ..................................................................................... 27
Educational Data Mining with KNIME .................................................................. 28
Educational Data Mining in Haaga-Helia UAS ..................................................... 29
5 Design ......................................................................................................................... 32
5.1 Visual architecture .............................................................................................. 32
5.2 Dataset ............................................................................................................... 32
6 Implementation............................................................................................................ 34
6.1 1st iteration .......................................................................................................... 34
6.2 2nd iteration ......................................................................................................... 35
7 Results ........................................................................................................................ 36
7.1 Academic withdrawal .......................................................................................... 36
 7.2 Avoidance orientation.......................................................................................... 37
7.3 Fear of failure ...................................................................................................... 38
7.4 School value ....................................................................................................... 39
7.5 Age vs avoidance orientation .............................................................................. 40
7.6 Nationality vs fear of failure ................................................................................. 40
7.7 Gender comparison ............................................................................................ 41
8 Discussion................................................................................................................... 43
8.1 Academic withdrawal .......................................................................................... 43
8.2 Avoidance orientation.......................................................................................... 43
8.3 Fear of failure ...................................................................................................... 44
8.4 School value ....................................................................................................... 45
8.5 Age vs avoidance orientation .............................................................................. 45
8.6 Nationality vs fear of failure ................................................................................. 45
8.7 Gender comparison ............................................................................................ 46
8.8 Recommendations .............................................................................................. 47
Conclusion .................................................................................................................. 48
9 48
References ...................................................................................................................... 49
Appendices ...................................................................................................................... 51
Appendix 1. Niemivirta questionnaire .......................................................................... 51
51

Table of figures

Figure 1 CRISP/DM method: Intelligent data analysis processing ......................................4

Figure 2 Related fields .......................................................................................................7
Figure 3 Population vs. Sample .........................................................................................8
Figure 4 Interquartile range..............................................................................................11
Figure 5 Positive Skew ....................................................................................................12
Figure 6 Normal distribution .............................................................................................12
Figure 7 Negative skew ...................................................................................................13
Figure 8 From data to knowledge ....................................................................................15
Figure 9 Machine Learning Processes .............................................................................19
Figure 10 Automatic Tools for Machine Learning & Data Mining ......................................21
Figure 11 KNIME Workspace Set Up ...............................................................................23
Figure 12 KNIME Introduction ..........................................................................................24
Figure 13 Node description ..............................................................................................25
Figure 14 KNIME workflow functions ...............................................................................25
Figure 15 KNIME project demo ........................................................................................26
Figure 16 EDM processes ...............................................................................................28
Figure 17 Lifecycle of data science ..................................................................................29
Figure 18 Organization of Haaga-Helia UAS....................................................................30
Figure 19 Visual architecture ...........................................................................................32
Figure 20 KNIME main workflow ......................................................................................34
Figure 21 Academic withdrawal comparison ....................................................................36
Figure 22 Avoidance orientation comparison ...................................................................37
Figure 23 Fear of failure comparison ...............................................................................38
Figure 24 School value comparison .................................................................................39
Figure 25 Age vs avoidance orientation ...........................................................................40
Figure 26 Nationality vs fear of failure ..............................................................................41
Figure 27 Gender comparison .........................................................................................42
Acknowledgements

I would like to start my thesis expressing my appreciation for all the support and time all
the lecturers have put during these years I have been studying in Haaga-Helia University
of Applied Science, in Helsinki.

First, I am very grateful to Dr. Amir Dirin that gave me the opportunity to be part of this
thesis project. He also has been a great thesis supervisor, arranging weekly meetings with
me for checking the progress. I am honoured to be his student and start my path in IT with
Dr. Dirin’s courses, going more professional with User Experience, Software Engineering
etc.

Pr. Dr Dominique Genoud for his amazing lectures in Data Mining. He has a fundamental
role to make me study more in Data Mining and Machine Learning. His knowledge in this
field is just inspirational and adorable.

Gjergji Make, student at Haaga-Helia UAS for cooperating with me during the thesis and
developing the KNIME project. We have had a lot of project together and a lot of learnings
supporting each-other.

Mr. Kari Silpiö, senior lecturer and my personal advisor during my studies in Haaga-Helia
UAS has had a great role in my study growth. Besides, he is an expert in database teach-
ing.

Mr. Juhani Välimäki, senior lecturer and the best programming teacher in Haaga-Helia
UAS. He has always been there to help me with every question I have had. Moreover, I
got a clear understanding of SCRUM methodology in just few lessons by Mr. Välimäki.

After all, I would like to thank every lecturer in Haaga-Helia University of Applied science
for all the support and contribution they have given to me for my professional growth. Ap-
preciation goes also for the whole personnel of the university.
Abbreviations

KNIME - Konstanz Information Miner

Weka - Waikato Environment for Knowledge Analysis
GUI - Graphical User Interface
EDM - Educational Data Mining
ML - Machine Learning
CRISP/DM - Cross-industry standard process for data mining
UAS – University of Applied Sciences
BIT – Business Information Technology
Td-idf – Term’s Frequency and Its inverse Document Frequency
1 Introduction

Data analysis is having a huge impact in today’s world. Through the years, we can notice
a huge development of data analysis especially from the evolution of computers.
Nowadays, we see an increasingly large importance of data analysis in all kind and size of
businesses with a lot of fancy and sophisticated tools and techniques available.

This thesis is about implementation of KNIME Analytics Platform in Haaga-Helia Univer-

sity of Applied Science to increase BITe student’s performance. By creating a KNIME
workflow, a detailed statistical analysis is available for future interpretations and predic-
tions.

KNIME Analytics Platform provides vast amount of analytics possibilities as well as predic-
tion capabilities. For better usage of these capabilities we need to keep in mind that the
dataset provided is accurate enough and has the required information. It is worth mention-
ing that the questionnaire conducted to Haaga-Helia UAS BITe students was based on
Niemivirta study of 8 scale factors of student performance.

The questionnaire was answered by 2018 students of BITe program. The structure was
based on numeric values for better analytics possibilities. There were about 100 students
who were able to answer. Later in the Implementation chapter we will explain deeper the
construction of the data collected from the questionnaire.

While doing this analysis, machine learning, data mining etc will be explained in detail, as
well as the technical information about KNIME Analytics Platform. For better results, we
must be aware for the current situation of the industry.

Data interpretation will play a big role in this thesis. Conclusions are generated based on
interpreting the charts and data provided by the platform. We will focus on comparing
main factors which influences the student performance the most such as fear of failure,
academic withdrawal or avoidance orientation factors. This will help in making conclusions
on what factors should be focused on to raise the performance

Finally, the reader will get a deep knowledge of the industry as well as how the overall
BITe students’ performance looks like in Haaga-Helia UAS.

1
2 Research Question

The main research question of this thesis is determining how we could assess the future
trends on BITe students’ performance through KNIME Analytics Platform.

2.1 Sub-Questions

Students’ performance is a psychology related topic that depends on internal, external

and natural factors. It explains how students progress or regress in an academic year and
aims to increase the performance through methodologies.

Educational Data Mining has significantly bloomed after the birth of the new technologies
which can control education related data. Several machine learning algorithms and data
mining and statistical techniques are used during educational data mining researches.
These researches aim to understand students’ way of learning in order to predict students’
academic performance and make improvements in the way students learn. (Roy & Garg,
2017)

Pointing to technology and considering the student as the main asset of the university, we
are now utilizing KNIME Analytics Platform to analyse students’ performance at Haaga-
Helia University of Applied Sciences in BITe degree program based on avoidance orienta-
tion, fear of failure, school value and academic withdrawal.

2
3 Methods

In this chapter, we are going to get an understanding of the methodology followed for writ-
ing this thesis, mixed methods and workflow processes.

3.1 Methodology

To fulfil the thesis and to achieve the best results wanted, we will divide the project in two
iterations. During the first iteration, we are going to develop the project in KNIME Analytics
Platform for proof of concept.

The second iteration is focused in interpretation of KNIME main workflow and discussion
of results.

3.2 Mixed Methods:

The research methodology that involves collection, analysing and integration quantitative,
through surveys, experiments, and qualitative, through interviews, focused groups etc, are
known as mixed methods research. The mixed methods research is used when the inte-
gration provides more effective understanding of the research problem than each of the
methods alone can provide. (Johnson, Onwuegbuzie, & Turner, 2007)

In this thesis we have collected data statistically, but we will have qualitative analysis. The
statistical data or quantitative data is collected through surveys to BITe students early in
2018. This includes all the close-ended information such as behaviours, students’ perfor-
mance, fear of failure etc. The qualitative data is aggregated into categories of data and
provides useful results.

3.3 Workflow Processes

It has a high importance the understanding of the process from the beginning until the de-
ployment of it. In this thesis, we are going to use CRISP/DM method because it provides a
structured approach for planning a data mining project.

CRISP/DM is a data mining method that provides an overview of the data mining project’s
life cycle. It contains all the phases of a data mining project, with their particular tasks. In
description level we can uncover some of the relationships between the tasks, while in all

3
 the other levels, all the relationships between the tasks are clear and intelligible. The rela-
tionships depends on the goals, interests of the user, background as well as the data pro-
vided, so they can be between any data mining tasks. (Ncr et al., 2000)

The phases of CRISP/DM method are as shown in the figure (figure 1).

Problem
Partly Understanding Partly

Cancel the
Project project
End Data
Deployment
Understanding

Yes Yes No
No

Data Pre-
Evaluation
Processing

Modelling

Improbable Probable

Figure 1 CRISP/DM method: Intelligent data analysis processing

Problem understanding

During the first phase we need to clarify what exactly is the problem we need to solve,
how should the solution look like and what are the knowns of the problem.

Improve tech-
Objective Data adapt
nical quality?
reached? to the problem?

4
The focus of this phase is the understanding of project objectives and requirements from a
business prospective. Then is important to convert all the knowledge into a data mining
problem so we can design a plan to successfully achieve this phase. (Chapman et al.,
2000)

Data understanding

It takes into consideration what available data we have, which is relevant, validity, quality
and sufficiency of the data related to the problem.

But how do we go through data understanding? First step we have to do is data collection.
Then we have to proceed with different activities for getting familiar with data provided.
Some of this activities are: identifying data quality problems, discovering first insights into
the data and detecting interesting subsets for making hypothesis regarding hidden infor-
mation. (Chapman et al., 2000)

Data pre-processing

We must remove the unnecessary data such as empty rows and rows with no meaning,
transform the data in the best way for modelling and increase data quality.

Data pre-processing phase is based on all the required activities that help to convert all
the initial raw data into a constructed final dataset. There are few tasks performed during
data pre-processing, such as table, record, attribute selection, as well as data transfor-
mation and cleaning for modelling tools. All these tasks can be performed several times
without following any order. (Chapman et al., 2000)

Modelling

We need to clarify the best model architecture that suits the problem, find out the tech-
niques for getting the desired model and best prediction accuracy.

Modelling phase includes a lot of selected and applied techniques. Their parameters are
calibrated to optimal value. For resolving the same data mining problem, we can use dif-
ferent modelling techniques. Some of the techniques have specific requirements on the
form of data, so it is necessary to go back in the data pre-processing phase. (Chapman et
al., 2000)

Evaluation

It is the phase we evaluate how good does the model meets the project’s requirements
and what have been our learnings. In case the goal is not met, we must end the project.

5
During modelling phase, we have already built the model that seems to be of a high qual-
ity from a data analysis perspective. Still, the model is not ready yet for deployment. That
is why we have evaluation phase in the middle. It is important to do a deep evaluation of
the model, review all the steps executed for creating it and make sure that the model
reaches the business objectives exactly as require. Also, we should ascertain that we
have adequately contemplated all the significant business issues that might raise. At the
end, we need to take a decision on the use of the data mining results that should be
reached. (Chapman et al., 2000)

Deployment

We need to understand how to deploy the project and the validity of the model for real
cases.

After we have created the model and evaluated it, we have the deployment phase. But
what is the deployment phase and what do we have to do? We already have the model
that has increased the knowledge of the data. This knowledge is easily understood by de-
velopers, but for it is highly important that the customer can understand and use it. The
most common way to visualize the data is on a real-time web page designed according to
customer’s needs.

On top of that, we can represent the knowledge of the data in a simple report. If required
by the customer, we can also implement a repeatable data mining process across the en-
terprise, which is a pretty complex implementation. The responsible person for deploy-
ment is, in most of the cases, the customer. When the data analyst takes care of deploy-
ment, he needs to make sure that the customer understand all the actions needed to be
carried out so he can profit from the models created. (Chapman et al., 2000)

6
4 Background Studies

Data Mining, Machine Learning, data science, etc are common words we see quite often
on news and they are considered “the hot topic” of technology. But what do all these
words mean and are they related with each other?

They are all related and subfields of one-another as shown in figure (figure 2). Deep
Learning is a subfield of Machine Learning, which itself is a subfield of Artificial Intelli-
gence. Computer science covers all these fields. A new term is Data Science that includes
ML and some computer science.

Computer
Science
AI
ML

Deep
Data Science
Learning

Figure 2 Related fields

A better understanding and explanation of the main terms we need in this thesis is given
below.

4.1 Statistical Analysis

Statistical analysis is a collection, examination, summarization, manipulation and interpre-

tation of quantitative data. It is used for uncovering trends, patterns, relationships and
causes that are inherent in the first sight. (WebFinance Inc., 2018)

Statistics are applied every day from the basic research, to companies and government.
From a vast amount of data, we try to manipulate it, so the data will have meaning. We

7
can summarize the data, calculate mean value and spread of data, make future predic-
tions and test hypothesis.

Statistical Analysis is a branch of mathematics dealing with data. It operates in two main
methods for data analysis: descriptive analysis and inferential statistics.

Descriptive analysis summarize data from a sample using parameters such as mean,
mode, standard deviation and median. It allows us to describe the data and it uses all the
data we are interested in or population to give us results.

Inferential statistics use sample data or only the part of the population we are interested to
investigate to draw conclusions and determine if the data can give us relevant prediction.
Population is the whole data we are interested in and sample is just the part of data we
need.

Sample

Population

Figure 3 Population vs. Sample

Statistical analysis methods

Once we have collected the data, we need to give meaning to them. In statistical data
analysis there are few methods that help us interpret the data collected.

Mean

Mean is widely known as the average. But why we use mean and how do we calculate it?
Mean is used to determine the overall trend, so we can have a snapshot of the whole data
provided. Calculating mean value is pretty easy and quick. We have to divide the sum of
the whole list of numbers with the number of the items in the list. (Pong, 2017)

8
Formula to calculate mean value is: x̄ = ( Σ xi ) / n, where x̄ is the mean, xi are all the x
values and n is the number of participants on the sample.

Standard deviation

Standard deviation evaluates the spread of data around the mean value. Sigma, the
Greek letter, stands for standard deviation. There are two types of standard deviation: low
and high. A low standard deviation means that more data is lined up with the mean value.
A high standard deviation foretokens that data is spread more widely from the mean
value. Besides, standard deviation is used for determining dispersion of data pints in
some data analysis methods. (Pong, 2017)

Formula to calculate standard deviation is: σ = √ (∑∣x− x̄∣²/ N), where σ stands for
standard deviation, x stands for a value in the data set, x̄ for mean value and n for number
of data in population.

Regression

Regression is a statistical process that shows the relationship between dependent varia-
bles and explanatory variables or independent ones which are used for prediction. There
are quite many techniques for making the models and analysis between variables and the
most common one is scatterplot. There we can check if we have strong relations between
the variables

Formula to calculate regression is: Y = a + bX, where Y stands for regression, a for the in-
tercept, b for slope of the line and X is explanatory variable.

Sample Size Determination

Sample size determination is used in all the cases we want to make some measures in a
population. For example, if we need to get some predictions about the elections in Hel-
sinki, we do not need to ask every inhabitant about their opinion, but we make assump-
tions and predictions based on the answers of a group of people. To make right predic-
tions, we need to define the right amount of people involved in the sample, and the correct
size of the sample is calculated by using standard deviation methods and proportion.

Hypothesis Testing

Hypothesis testing is all about making assumptions in a defined parameter of a population

and it finds a widely usage in researches, science, businesses etc. Experimental data is
used for making statistical decisions. Hypothesis testing is considered to be relevant when
the results we get are not happening by chance.

9
Statistical analysis techniques

There are few techniques that help us to make statistical analysis, such as: summarizing
data, measuring the location of data, measuring the spread of the data and skew. A de-
scription of each technique is given below.

Summarizing Data: Grouping and Visualizing

Data summarizing is a combination of grouping and visualization. First thing to start with is
grouping the raw data into categories and then visualize it. For example, taking this thesis
into consideration, we are interested into differences by gender with the eight Niemivirta
categories (appendixes 1). This way we have two groups: male and female for making
comparison. To visualize the results we get, we use can bar charts, graphs etc.

Measures of Location: Averages

Average is a calculated central value of a group of numbers. It gives us information of the

size of the effect we are trying to test. In other words, it is the division of the sum of all
numbers we have with the amount of the numbers in the list. There are three measure of
average: mean, mode and median. Habitually, when people want to express something
about average, they refer to mean value. Taking in consideration all these three types of
average, they all differ from one another, so we cannot consider all them the same. It is
very important to clear out for which one we are referring. (Skillsyouneed, 2017)

Average uses all the acquired data values for statistical analysis but can be skewed by ex-
treme values. To avoid this situation, researchers utilize median instead. Median is the
mid-point of all the data and it does not get skewed by outliers. Regardless, median is not
very accurate for further statistical analysis. Mode present the most common value in a
data set, but it is not unerring enough for statistical analysis. (Skillsyouneed, 2017)

Measures of Spread: Range, Variance and Standard Deviation

Researchers often want to look at the spread of the data, that is, how widely the data are
spread across the whole possible measurement scale.

To measure the spread of the data we can use range, variance and standard deviation. A
brief description of these three terms is given below.

Range is the difference between the largest values with the smallest ones. Ofttimes we
can hear the interquartile range used by researchers. But what is it? Interquartile range is

10
the range of the middle half of the data, from 25% of the lower quartile to the upper quar-
tile up to 75%, of the values. To find the value of the quartiles, we follow the same process
as we do with median. Median is all the time 50% of the values. Upper quartile is 75% of
the values and lower quartile is 25% of the values. (Skillsyouneed, 2017)

For a easier understanding of the interquartile range, we can use figure (figure 4).

Figure 4 Interquartile range

Standard deviation measures the average spread of the data. A more detailed description
can be found in chapter 4.1.1.2

The variance is the square of the standard deviation. They are calculated by:
1. Calculating the difference of each value from the mean
2. Squaring each one (to eliminate any difference between those above and below
the mean)
3. Summing the squared differences
4. Dividing by the number of items minus one.
This gives the variance.

Standard deviation is the square root of the variance.

Skew

Skew is used to evaluate how symmetric is the data set visualized into a graph, which
means it demonstrates if we have more high or low values. The skew can be positive,
negative or we have no skew at all. When we have high values, then the skew is positive
and when we have low values, the skew is negative. There is no skew when the distribu-
tion between high and low values is the same. A high skew indicates a low mean, mode
and median.

Let’s understand the negative, neutral and positive skew.

In the figure (figure 5), we have a positive skew. In a positive skew, the tallest line is al-
ways on the left side of the graph, as well as the mean value is on the left.

11
 Positive Skew
A B C D E F G H I J

14
12
10
8
6
4
2
0

Figure 5 Positive Skew

There is no skew when there is a normal distribution which creates a symmetrical and
identical graph in both sides. The mean, mode and median are exactly at the peak of the
graph, as shown at the figure (figure 6).

No skew
A B C D E F G H I J K

14

12

10

Figure 6 Normal distribution

There is a negative skew, when the tallest line of the graph is on the right side it as shown
in the figure (figure 7). Also, the mean value can be found on the right side of the graph.

12
 Negative Skew
A B C D E F G H I J

14

12

10

Figure 7 Negative skew

4.2 Data Mining

Recently we can notice a tremendous number of digital data due to the advancing tech-
nologies. In this case, it was required a scientific way to extract the useful information from
the huge data repositories. There are few explanations for data mining. Below you will find
two definitions from Margaret Rouse and SAS Institute.

In enterprises, we need to solve problems through data analysis and we have in front of
us a large data set where we need to sort the data for recognizing patterns and setting up
relationships. All this process is done through data mining. Tools used in data mining give
the opportunity to enterprises to make future trends predictions. (Margaret Rouse, 2017)

In a large data set where we need to predict outcomes, we use data mining process. That
allows us to find anomalies, patterns and correlations. The wide variety of different tech-
niques, allows us to use the information for increasing revenues, cutting costs, reducing
risks, improving customer relationships etc. (SAS Institute, 2018)

Data mining in years and now

“Knowledge discovery in database” is the term used decades ago for referring data min-
ing. This new term, “data mining” came in use in 1960s and now is a key word in technol-
ogy. Data mining is the process of uncovering unknown relations in the data and making
future trends predictions. It is always evolving as the amount of data is growing so fast.
(SAS Institute, 2018)

13
Data mining is an entwined settlement of three scientific disciplines: statistics, artificial in-
telligence and machine learning. Statistics is used for making numeric study of data rela-
tionships, artificial intelligence shows human-like behaviours displayed by the software
and machine learning uses algorithms that learn from current data for making future pre-
dictions. (SAS Institute, 2018)

In the last decade, especially in the recent years, we can notice a radical enhancement of
technology. Many machines and processes have been automated. Within the same time
limit, we can be more productive than before. Besides, the work and energy required for
the same job is relatively less than it used to be.

Another important thing about this upgraded technology is related with data complexity. A
complex data set gives us a wider view of insights. That is why businesses, banks, manu-
factures, retailers etc use data mining to lighten up relationships from social media, risk,
revenue, pricing etc.

Importance of Data Mining

The amount of data is growing rapidly and it is doubling every other year. The greatest
part of the digital universe is formed by unstructured data. Be that as it may, more data
does not indicate more knowledge. So, why is data mining important?

Data mining allows you to:

1. Sift through all the chaotic and repetitive noise in your data.
2. Understand what information is relevant and later use it for making likely out-
comes.
3. Speed up the rhythm of taking decisions
(SAS Institute, 2018)

Data Mining Methods

There is a huge number of Data Mining Methods, but they are categorized in two major
ones. We have Descriptive methods and Predictive methods. In this thesis we are going
to use descriptive methods.

The Descriptive methods, analyse the data and check for interpretable patterns to de-
scribe the data but don’t do any mathematical analysis. The Predictive methods use varia-
ble for predicting future outcomes.

14
How does Data Mining work?

To understand the process, we start from raw data to obtain the knowledge. As shown in
figure (figure 8), the steps to be followed are:

Data → Selection → Pre-processing → Transformation → Data Mining → Knowledge

During the Selection phase, we go from a bunch of raw data to the target data, by remov-
ing non-useful data or data that don’t fit our main format. From the target data we go to
pre-processed data by detecting outliers and missing values. After pre-processed data,
we go to transformed data by normalizing it or finding correlated variables. The trans-
formed data helps us to create patterns by using several data mining algorithms. Then,
the user can create the knowledge by interpreting the patterns.

Data Mining Knowledge

Transformation

Pre-processing

Selection

Data

Figure 8 From data to knowledge

4.3 Machine Learning

Machine Learning is a system that is continuously improving and becoming more unerring
in predictions with the more data we annex.

It is a very wide topic with lots of discussions and definitions. It is considered as a new
topic in technology, but it lies years back. Roots of Machine Learning are in statistics, the
art of extracting knowledge from data.

15
History of Machine Learning

Machine leaning’s history foredates the computers, but it is still considered as an excres-
cence of statistics.

A brief history of machine learning is given below.

1943 - 1959 Neural Network

In 1943 where a human "neural network" is modeled with electrical circuits. This was used
by computer scientists in 1950s. This is the point where Machine Learning started to be-
come reality. In 1959, the phone calls became clearer due to implementation of adaptive-
filter.(Orcibal & Jerphagnon, 2018)

Predictive analysis:
Anderson is one of the first ones used predictive analysis. He observed three different
types of Irises and documented all the measurements. By using those measurements, An-
derson developed a simple algorithm that could predict the species of the flower. The da-
taset for this algorithm was pretty small and could haven been solved effortlessly by hand.
But the idea of all this was to use a wide range of methods. Algorithm based on neural
networks are different from the methods that are used for discovering potential combina-
tions of rules that execute the classification. (Houle, 2016)

1957 Perceptron:
The most straightforward neural network is Perceptron. Perceptron was a vogue in 1960s.
It acquired knowledge to a separating knowledge between two different categories. It was
operating as a parallel computer and could perform much better than conventional com-
puters at visual recognition tasks. In 1969, Minsky and Papert published a book that
proved the limitations of Perceptron. This is the year, Perceptron vanished hastily. (Houle,
2016)

1971 Full-text search:

In 1971, we have the development of the algorithm for full-text search. This is known as
Term Frequency (TF) and Inverse Document Frequency (IDF) or tf-idf algorithm. This al-
gorithm is pretty similar with Perceptron. Tf-idf is a computation of a dot product in a high-
dimensional space. The document we are using is considered a list of scores based on
how the word is used. Then we can compute the similarity metric between two documents
or one document and one query.

16
How to understand this? We have a document with 1000 words and we have to find Tf
and Idf for the term “algorithm” which has been used 50 times. Also, the term “algorithm
has been used Z amount of times in a 200 000 000 principal and in 100 000 documents
we have involved the term “algorithm”. Tf will show us the frequency of the word “algo-
rithm”. Idf will show how noteworthy the term “algorithm” is in the whole principal.

To calculate the tf we have the formula:

TFterm = Nterm / Nwords

In our case we have:

TFalgorithm = Nalgorithm / Nwords = 50 / 1000 = 0.05

To calculate idf we have the formula:

IDFterm = log (NDF / Nwords)
In our case we have:
IDFterm = log (200 000 000 / 100 000) = log (200) = 2.3

1992 US post office: Handwritten digits:

Neural networks had continuous progress from the beginning till 1992. It was put a lot of
importance to visual processing in the human brain. The researchers tracked down that
human brain is built from several layers. These layers encode all the multiplex features of
the images. This result had a fundamental role for further development, even thought it
took time to come up with the ideal method for training the neural networks. However,
when this ideal method was developed, the multiple-layer neural networks had an essen-
tial and effective role in visual recognition tasks. In 1992 the US Post Office adapted an
automatic ZIP code reading machine which was developed as a result of the multiple-
layer neural networks.

Neural networks are trained to memorize different cases and examples and this is the way
they operate. Presuming that the examples we use for training the neural networks are
just basic ones, this will not be efficient. A neural network with a single hidden layer can
learn any function and if we add more layers we can form “information bottlenecks” that
help in sorting the information. (Houle, 2016)

Data driven competitions:

To understand the data driven competition we can take as an example the Fisher-Ander-
son Iris dataset. There are many scientists and programmers who have been working
hard to ameliorate the performance of a task averse to standardized data, indeed it can

17
never be an overvaluation of it. Quite many improvements in image recognition has came
as a result of competitions around ImageNet. (Houle, 2016)

2006 Deep learning:

Deep learning is a recent concept in the world of technology, that was initiated by Geoff
Hinton and his students. This concept is related with multiple-layer neural networks and
the way of training these neural networks. To simplify the process, Geoff and the students
decided to divide it in two phases. During the first phase, the network is handled the same
way as a Bayesian Belief Network. This network was instructed to uncover statistical regu-
larities in the data. To make the input data more accessible for further usage, they chose
not to label it. In this way, if the model was shown a face, it develops a “theory of faces”.
This would give model a chance to be more open to important discoveries related with
face’s features, and running the network backwards, enables the model to draw faces and
face-like-shapes. Indeed, first phase creates a model of inputs that can be used during the
second phase. In the second phase all the information is used for training multiple hidden
layers. The data is now labeled, that can help for classifying the images based on the fea-
tures. (Houle, 2016)

You can find more about deep learning in chapter 4.4.

Present: Explosion of Neural Network architectures:

Nowadays, machine learning has been ameliorated quite a lot and neural networks are
just a small part of it. Still and all, they are very active. A lot work has been done to organ-
ize deep networks for a more proficient recognition of image and speech. One area of
neural network is recurrent neural network. This finds usage in text understanding and ar-
eas where a stream of symbol should be transformed into an object. It can deal with sev-
eral inputs happening as a series in time. (Houle, 2016)

What is Machine Learning?

“Machine learning is the science of getting computers to act without being explicitly pro-
grammed.” (Pyle & San Jose, 2015)

The basic of machine learning starts with the practice of using algorithms to parse data,
learn from that data and later start making predictions about different things in the world.
(Copeland, 2016)

18
On the other side, machine learning is an answer for the question “How can we build com-
puter systems that automatically improve with experience, and what are the fundamental
laws that govern all learning processes?” (Mitchell, 2006)

In other words, Machine Learning is the science of teaching computers to act like humans
but getting more productive and predictive than us. We offer them data from real world, so
the processes of algorithm are improved.

Nowadays, many application use Machine Learning to personalize members’ data for fu-
ture personalized suggestions. Some of the most well-known applications that use Ma-
chine Learning are: Facebook News Feed, Netflix, Snapchat, Google Maps, Tinder etc.

But how does Machine learning works? The processes that need to be followed are
shown in the figure (figure 9).

Identify Identify ML Build

relevant analytical Train the Run the
algorithm model model
dataset for to be used model
analysing

Figure 9 Machine Learning Processes

Types of Machine Learning

Machine Learning functions in three different algorithm types: supervised learning, unsu-
pervised learning and reinforcement learning.

Supervised learning

The supervised learning consists in a given input predicting the correct output. The train-
ing of this algorithm continues until we get the desired result.

In simpler words, if we have a picture with a number on it, the machine learning must pre-
dict what is the right number on it. Human does not write down rules for classification, but
it trains the learner to recognize the correct output automatically, by providing correct an-
swer.

19
Decision tree, regression etc are part of supervised learning.

How it works?

We have an outcome variable, or differently known as the dependent variable, that is used
to be predicted by a given set of predictors or independent variable. These creates the al-
gorithm. The set of these variables is used for generating a specific function for mapping
inputs with the desired outputs. This is the training process and it continues until the
model achieves the desired level of accuracy on the training data. (Sunil, 2016)

Unsupervised learning

In unsupervised learning humans do not provide the correct answer, so the process is a
bit more complicated. The learner tries to get some data from visualization, clustering or
generative modelling.

How it works?

During unsupervised learning there is not provided an outcome variable to predict. This
prediction is used for clustering population in different groups. That is why it finds a wide
usage in segmentation of customers in different groups for specific intervention. (Sunil,
2016)

Reinforcement learning

Reinforcement learning is used in more sensitive cases, such as self-driving cars. Contin-
ues feedback is given to the learner, so the result is more accurate.

How it works

In reinforcement learning algorithm, the machine is trained to make distinct decisions. The
machine learns continuously from the previous experience. First, we expose the machine
to an environment. Then, the machine should train itself using trials and errors. Replica-
tion of the process, gives the machine the possibility to improve ceaselessly until it cap-
tures the best knowledge for making business decisions. (Sunil, 2016)

20
Automatic Tools for Machine Learning and Data Mining

A variety of tools are used in analysis statistical data and data mining. Data mining tools
are divided in three categories: free software, free software with paid service and enter-
prise software. A list of most used software is presented in the figure (figure 10).

Automatic Tools for

ML and Data Mining

Frees softwares with Enterprise Data

Free softwares
paid service Mining softwares

R KNIME SPSS Modeler

Weka RapidMiner SAS Enterprise Miner

Cortana

Python

Figure 10 Automatic Tools for Machine Learning & Data Mining

Machine Learning and Data Mining main tools in years

1993 - WEKA:
Weka is a data mining and machine learning toolkit with an extensive usage nowadays. It
was initially developed at the University of Waikato in New Zealand and now it is quite
well-known to professors, universities and industrial and academic researchers. Weka
uses Java for writing data mining algorithms and it has a huge collection of state-of-the-art
machine learning algorithms. Moreover, regression, clustering, classification, visualization,
association rules and data pre-processing are supported by Weka. That is why Weka is a
very powerful tool. (Naik & Samant, 2016)

Weka is an open source Machine Learning software that does not require any program-
ming skills since it has a GUI. There are three ways Weka operates:

21
 1. Explorer – allows people to play with data and transform it according to desired al-
gorithms
2. Experimenter – allows people to analyse the results of the algorithm chosen
3. Knowledge Flow – allows people to create the graphic design of the process and
run it

1997 - Orange:
Orange is another open source data mining and machine learning platform, developed
and maintained by the Bioinformatics Laboratory of the Faculty of Computer and Infor-
mation Science at University of Ljubljana. It is written in Python and it provides visual pro-
gramming front-end for visualization and investigative data analysis. (Naik & Samant,
2016)

Orange has a canvas interface with widgets that vary from the simplest data visualization
to predictive modelling. Workflows can be created through user-designed widgets or Py-
thon libraries.

2001 - RapidMiner: is a user interactive environment for machine learning and data min-
ing processes. It is opensource, free project implemented in Java. It represents a modular
approach to design even very complex problems - a modular operator concept which al-
lows the design of complex nested operator chains for a huge number of learning prob-
lems.

2003 - Tanagra:
Tangara is a free suite of machine learning software. It was developed at the Lumiere Uni-
versity Lyon 2, in France by Ricco Rakotomalala. It gets usage in researches and aca-
demic programs. Similar to Weka, quite many data mining tasks are supported by Tan-
gara such as classification, clustering, regression, visualization, factor analysis, instance
selection, feature selection, feature construction, descriptive statistics and association rule
learning. (Naik & Samant, 2016)

Tanagra is a simplified data mining tool that works with diagrams and nodes. The models
are shown in a tree diagram and the results are displayed in HTML format.

2004 - KNIME:
KNIME is an open source data mining and visualization platform. It is used worldwide by
more than 3000 organizations. It was developed at the University of Konstanz. It is based
on Eclipse IDE platform, making it a quite powerful development and data mining platform.
KNIME desktop in the entry version of KNIME. (Naik & Samant, 2016)

22
KNIME is a powerful and simple platform that allows everybody to create machine learn-
ing solutions on a GUI based workflow. No coding skills are required. This thesis is based
on KNIME Analytics Platform, so we will get to know a lot about it.

KNIME Analytics Platform

According to chapter 4.5, KNIME is a powerful platform for Data Mining and Machine
Learning solutions. In this section, we will have a step-by-step introduction to KNIME Ana-
lytics Platform, so at the end the user will know how to build a project.

Installation

First, we download KNIME from the official web page: http://knime.com

When the installation is ready, we set up our workspace, as shown in the figure (figure
11).

Figure 11 KNIME Workspace Set Up

First Introduction with KNIME

After the installation, we are in contact with the view of the platform as shown in figure (fig-
ure 12).

23
Figure 12 KNIME Introduction

We have the menu bar on top of the page: File, Edit, View, Help. We can update the plat-
form under File -> Update KNIME and then Restart.

In KNIME Explorer, it is the workspace repository that can be in server or local.

In KNIME Repository, there are located all the nodes available in KNIME platform, so it is
a node repository.

Welcome to KNIME Analytic Platform is the workflow development space where we can
create our project.

KNIME Console is an error and information console.

Introduction to software

KNIME is a very simple platform that works on a GUI based workflow, functioning with
nodes. Nodes are divided into colors, where each color represents a function. Orange
color is used to nodes that read information and manipulate it. Writing nodes are in red,
visualizing nodes are blue, mining nodes are green etc.

24
For adding or replacing nodes, adding files, we can just drag and drop. Another way to
add nodes, is by double-clicking on their name in the Node Repository. To check the
node’s configuration, we need to double-click on it.

In case of confusion about nodes’ function, we can click on the node and check node de-
scription as shown in figure (figure 13).

Figure 13 Node description

In workflow, there are six main functions as represented in the figure (figure 14).

1. Execute selected nodes

2. Execute all
3. Execute and open first view
4. Reset selected nodes
5. View results of selected nodes
6. View steps

1 2 3 4 5 6

Figure 14 KNIME workflow functions

A demo of a small project is presented in the figure (figure 15). It shows 4 stages that we
need to follow: Read, Transform, Analyze and Deploy.

25
 Read Transform Analyze Deploy

Figure 15 KNIME project demo

As shown in the figure (figure 14), we have used file reader that reads excel, csv file etc.,
partitioning, decision tree learner, decision tree predictor and scatter plot.

File reader: “This node can be used to read data from an ASCII file or URL location. It can
be configured to read various formats.” (KNIME.COM AG, 2016)

Partitioning: “The input table is split into two partitions (i.e. row-wise), e.g. train and test
data. The two partitions are available at the two output ports.” (KNIME.COM AG, 2016)

Decision tree learner: “This node induces a classification decision tree in main memory.
The target attribute must be nominal. The other attributes used for decision making can be
either nominal or numerical. Numeric splits are always binary (two outcomes), dividing the
domain in two partitions at a given split point. Nominal splits can be either binary (two out-
comes) or they can have as many outcomes as nominal values. In the case of a binary
split the nominal values are divided into two subsets. The algorithm provides two quality
measures for split calculation; the gain index and the gain ratio.” (KNIME.COM AG, 2016)

Decision tree predictor: “This node uses an existing decision tree (passed in through the
model port) to predict the class value for new patterns.” (KNIME.COM AG, 2016)

Scatter plot: “Creates a scatterplot of two selectable attributes. Then each datapoint is dis-
played as a dot at its corresponding place, dependent on it's values of the selected attrib-
utes. The dots are displayed in the color defined by the Color Manager, the size defined
by the Size Manager, and the shape defined by the Shape Manager.” (KNIME.COM AG,
2016)

26
4.4 Deep Learning

“The field of artificial intelligence is essentially when machines can do tasks that typically
require human intelligence. It encompasses machine learning, where machines can learn
by experience and acquire skills without human involvement. Deep learning is a subset of
machine learning where artificial neural networks, algorithms inspired by the human brain,
learn from large amounts of data. Similarly, to how we learn from experience, the deep
learning algorithm would perform a task repeatedly, each time tweaking it a little to im-
prove the outcome. We refer to ‘deep learning’ because the neural networks have various
(deep) layers that enable learning. Just about any problem that requires “thought” to figure
out is a problem deep learning can learn to solve. Deep learning allows machines to solve
complex problems even when using a data set that is very diverse, unstructured and inter-
connected. The more deep learning algorithms learn, the better they perform.” (Marr,
2018)

Educational Data Mining

Applying Data Mining in educational context is an emerging topic and it is known as EDM
(Educational Data Mining). EDM is the science of applying Data Mining and Machine
Learning in educational institutes aiming to increase study methods for improving stu-
dents’ performance by predicting future learning behaviour.

“Educational Data Mining (EDM) is an emerging discipline, concerned with developing

methods for exploring the unique types of data that come from educational settings, and
using those methods to better understand students, and the settings which they learn
in.”(Team, 2013)

“Educational data mining is emerging as a research area with a suite of computational and
psychological methods and research approaches for understanding how students learn.
New computer-supported interactive learning methods and tools—intelligent tutoring sys-
tems, simulations, games—have opened up opportunities to collect and analyse student
data, to discover patterns and trends in those data, and to make new discoveries and test
hypotheses about how students learn. Data collected from online learning systems can be
aggregated over large numbers of students and can contain many variables that data min-
ing algorithms can explore for model building.” (Team, 2013)

In other words, EDM is collecting data from students and analyse those to check for risk of
failing in final exams or students who want to drop university or degree program. This will

27
identify the students and give recommendations for future improvements. So EDM is an
educational software that predicts outcome.

Educational Data Mining Processes

The processes that are followed during Educational Data Mining are shown in figure (fig-
ure 16)

Data
collection

Improve
Data
learning
analysis
(Learner)

Data post-
Feedback
processing

Figure 16 EDM processes

The EDM starts with data collection by conducting a survey to the students of BITe degree
program in Haaga-Helia UAS. We go through all the data collected and start analyzing it.
After, we must pre-process the data, feedback it and improve learning.

This is a cycle process, so every time a new data is added in the database, the process
starts over again.

Educational Data Mining with KNIME

In this thesis, we are going to make students’ performance analysis and some future pre-
dictions based on past behaviours. The Data Mining software we are using is KNIME Ana-
lytics Platform.

To do some data mining analysis in educational context using KNIME Analytics Platform,
we need to follow a lifecycle as shown in figure (figure 17).

28
 Data Feature
Collect data Clean data
integration extraction

Model Model Predictive Feature

evaluation deployment modeling selection

Figure 17 Lifecycle of data science

The main steps followed to get some analytic results and predictions are described in
chapters 5 and 6.

Educational Data Mining in Haaga-Helia UAS

Haaga-Helia UAS is the university we are implementing educational data mining through
KNIME Analytics Platform. Before we go to the implementation of KNIME Analytics Plat-
form, there is an introduction of this university.

Haaga-Helia UAS

Haaga-Helia is a privately-run university of applied science, part of the Finnish public edu-
cational system, steered and co-founded by the Finnish Ministry of Education. (Haaga-
Helia UAS, 2018b)

In figure (figure 18), there is a visual organization of Haaga-Helia UAS.

29
 Financial
Service

HR Service

School of
Experience Digital IT Services
Business Vocational
& Wellness Business & Corporate
Programs Teacher Ed-
Economy Planning
ucation

Research, Development & Innovation Services

Higher
Commercial Services Education
Services
Open UAS

President, Managing Director

Board of Haaga-Helia University

of Applied Sciences Oy Ltd.

Figure 18 Organization of Haaga-Helia UAS

Haaga-Helia UAS provides quite many bachelor’s degree programs, but the focus for us is
the degree in Business Information Technology

Business Information Technology degree program

Business Information Technology degree program provides students with a wide range of
professional skills in both business and information technology. This degree offers a com-
bination of practical and extensive theoretical base, as well as practical application in co-
operation with It or other kind of companies, both locally and internationally. This co-oper-
ation provides the students the opportunity to learn new skills and acquire the experience
of working in a multicultural environment and on multinational projects. This is why, the in-
tegral part of BITe degree program is the collaboration with these companies. (Haaga-
Helia UAS, 2018a)

Another important sector of this degree program is the career planning. This includes the
development of entrepreneurship, sales and service abilities. What is more, there are
many courses enhancing team work, communication, ability to work independently, skills
required for being successful in a business environment. (Haaga-Helia UAS, 2018a)

So, as we can see, Business Information Technology degree program, enables the stu-
dents everything needed for being successful in the career path.

30
EDM in BIT degree program

This thesis is conducted by a Haaga-Helia BITe student, so the knowledge about this de-
gree program and experience are relevant for further research.

With implementation of educational data mining in BIT program, we aim to increase stu-
dents’ interest towards courses, how to minimize students’ withdrawal and find and effi-
cient way to match students’ interests and strong points with the relevant courses.

Educational data mining is very beneficial for the university as well, because it will have
more graduated students within three and a half years (degree program study timeline).

31
 5 Design

This part will describe briefly the visual architecture of the whole process from data collec-
tion till visualization and the dataset of the project. More detailed information can be found
in Gjergji Make’s bachelor’s thesis.

5.1 Visual architecture

As shown in figure (figure 19), the first step was to distribute the questionnaire to BITe
Haaga-Helia students. When the questionnaire is filled, we are provided an Excel data-
base with all the information. It is important to have data understanding and pre-pro-
cessing, where we can filter the data provided and remove unnecessary data, such as
wrong data.

We use this database to generate a machine learning database in KNIME Analytics Plat-
form and start building the project. We create different meta-nodes and/or wrapped nodes
for different data visualization.

The final step is data visualization and interpretation.

Data collection Data manipulation

Data visualization
from the question- and build of pro-
and interpretation
naire ject on KNIME

Figure 19 Visual architecture

5.2 Dataset

The dataset we are working during the project are the results gathered from the question-
naire (NIEMIVIRTA M. 2002) distributed to Haaga-Helia UAS students of BITe degree
program in 2018. The questionnaire has 30 questions. The first 6 questions reveal basic
information about the students and they are the main reference for further statistical analy-
sis interpretations.

The dataset generated from the variables of the first 6 questions are as follow:

1. Gender:
1. Female
2. Male

32
2. Age:
1. 18 – 21 years old
2. 22 – 25 years old
3. 26 – 29 years old
4. 30 – 35 years old
5. Older than 35 years

3. Nationality:
1. Finnish
2. European
3. South American
4. North American
5. Asian
6. African

4. Study Semester
1. First semester
2. Second semester
3. Third semester
4. Forth semester
5. Fifth semester
6. Sixth semester
7. Seventh semester
8. Etc

5. Accomplished a higher education before or not

6. BIT program was first choice or not

The other 24 questions are grouped in 8 categories with three questions each and scaled
from 1 to 7 by the students.

The 8 categories according to (NIEMIVIRTA, 2002) are:

1. “Mastery-intrinsic orientation: “to acquire new knowledge is an important goal for
me at school.
2. Mastery-“extrinsic orientation: my goal is to succeed in school.
3. Performance-approach orientation: “an important goal for me at school is to do
better than other students.
4. Performance-avoidance orientation: “I try to avoid situations in which I might fail or
make mistakes.
5. Avoidance orientation: “I try to get away with as little effort as possible in my
school work.
6. Fear of failure: “during classes or exams I often worry that I do worse than the
other students.
7. Academic withdrawal: “refers to an individual’s tendency to give up or withdraw
from demanding or difficult learning or performance situation.
8. School value” (NIEMIVIRTA, 2002)

The detailed questionnaire is available in Appendixes 1.

33
6 Implementation

As indicated in chapter 3.1, we will work with 2 iterations. In the first iteration we will focus
in the proof of concept to the deployment of the project. The second iteration we will focus
on interpretation of KNIME workflow and discussion of results.

6.1 1st iteration

The first iteration is divided in six phases based on the CRISP/DM method. Business un-
derstanding is done in cooperation with Mr. Dirin and Gjergji Make, where the current sta-
tus of Haaga-Helia UAS students is analysed and some alternative solutions have come
into consideration. Also, the questionnaire (NIEMIVIRTA M. 2002) is handled to students
of BIT degree program for data collection.

After that, together with Gjergji Make, we have followed the other five phases of
CRISP/DM method, using KNIME Analytics Platform for the optimal solution. The big pic-
ture of KNIME workflow is as shown in figure (figure 20).

Figure 20 KNIME main workflow

In the KNIME main workflow we have Data Manipulation meta-node where all the data
manipulations are handled. Also, there are wrapped meta-nodes. Each wrapped meta-

34
node is a group of JavaScript visualization nodes that can be displayed in a single HTML
page.

Please read Gjergji Make’s bachelor’s thesis for a deep understanding of KNIME workflow
processes that have been followed to achieve this data visualizations and interpretations
of this thesis.

6.2 2nd iteration

During the second iteration of this thesis is about results, recommendations and discus-
sions. These can be found in the upcoming chapters.

35
7 Results

Chapter 7 is dedicated to give straight results from the main bar charts created from all
the data analysis that have been done during this thesis-projects.

In chapter 8, there is interpretation of results, so the graphs have meaning.

Some conclusions and recommendations are given on chapter 9.

7.1 Academic withdrawal

As shown in the figure (figure 21), academic withdrawal is affecting the average factor of
mastery-intrinsic orientation, mastery-extrinsic orientation, performance approach-orienta-
tion and performance-avoidance orientation.

An average factor of 6.5 of mastery-intrinsic orientation is when we have a low academic

withdrawal. The average factor of mastery-intrinsic orientation drops to 6.4 when there is
medium academic withdrawal and 6.0 when there is high academic withdrawal.

An average factor of 5.7 of mastery-extrinsic orientation is when we have low academic

withdrawal. When the academic withdrawal is medium, the average factor raises up to 5.8
and it drops to 5.4 when the academic withdrawal is high.

Figure 21 Academic withdrawal comparison

36
The performance-approach orientation has an average factor of 4.0 when the academic
withdrawal is low. It raises up to 4.1 when academic withdrawal is medium and drops to
3.8 when the academic withdrawal is high. But as it is noticed from the graphic, the differ-
ence between the average factors is quite low.

An average factor of 2.8 of performance-avoidance orientation is when we have low aca-

demic withdrawal. When we have a medium academic withdrawal, the average factor of
performance-avoidance orientation raises up to 4.3 and to 4.5 when the academic with-
drawal is high. So, a low academic withdrawal indicates a higher average factor of perfor-
mance-avoidance orientation.

7.2 Avoidance orientation

As shown in the figure (figure 22), an average avoidance orientation indicates a lower
mastery-intrinsic orientation average factor. An average factor of 6.6 of mastery-intrinsic
orientation is when we have a low avoidance orientation. The average factor drops to 6.2
when the avoidance orientation is medium and raises up to 6.7 when the avoidance orien-
tation is high.

An average factor of 6.1 of mastery-extrinsic is when we have low avoidance orientation.

The average factor reduces to 5.6 when the avoidance orientation is medium and to 5.0
when the avoidance orientation is high. So, the higher the avoidance orientation, the lower
the average factor of mastery-extrinsic orientation is.

Figure 22 Avoidance orientation comparison

37
The performance-approach orientation has an average factor of 3.8 when the avoidance
orientation is low. The average factor of it raises up to 4.5 when the avoidance orientation
is medium and drops to 3.2 when the avoidance orientation is high. So, the average factor
of the performance-approach orientation gets higher as the avoidance orientation is me-
dium.

An average factor of 3.0 of performance-avoidance orientation is when we have low

avoidance orientation. It raises up to 4.2 when we have a medium avoidance orientation
and drops to 3.7 when the avoidance orientation is high.

7.3 Fear of failure

As shown in figure (figure 23), fear of failure is not affecting the mastery-intrinsic orienta-
tion average factor, but it has affected the mastery-extrinsic orientation average factor,
performance-approach orientation average factor and performance-avoidance orientation
average factor.

An average factor of 6.4 of mastery-intrinsic orientation is when we have high, medium or

low fear of failure.

An average factor of 5.4 of mastery-extrinsic orientation is when we have low fear of fail-
ure. When we have medium fear of failure, the average factor of mastery-extrinsic orienta-
tion raises up to 5.8, and 6.2 when the fear of failure is high. So, the higher the fear of fail-
ure is, the higher the mastery-extrinsic orientation becomes.

Figure 23 Fear of failure comparison

38
An average factor of 4.4 of performance-approach orientation is when the fear of failure is
high. When we have medium fear of failure, the average factor of performance-approach
orientation drops to 4.2 and to 3.5 when the fear of failure is low. As the fear of failure gets
lower the performance-approach orientation drops.

An average factor of 5 of performance-avoidance orientation is when we have a high fear

of failure. When we have medium fear of failure, the average-factor of performance-avoid-
ance performance drops to 3.7 and to 2.6 when the fear of failure is low. So, the higher
the fear of failure, the higher gets the performance-avoidance orientation average factor.

7.4 School value

As shown in figure (figure 24), school value has a huge impact in mastery-intrinsic orienta-
tion, mastery-extrinsic orientation and performance-avoidance orientation. The effect of
school value in performance-approach orientation average factor is almost null.

An average factor of 6.5 of mastery-intrinsic orientation is when we have high school

value. When we have a medium school value, the average factor of mastery-intrinsic
drops to 6.2. A low school value will drop the mastery-intrinsic orientation average factor
to 3.7.

Figure 24 School value comparison

An average factor of 4.0 performance-approach orientation is in all the cases remaining

the same.

39
An average factor of 3.6 of performance-avoidance orientation is when we have high
school value. The average factor raises up to 3.9 when the school value is medium and
4.3 when the school value is low. So, as the school value is going down, the performance-
avoidance orientation average factor is raising up.

7.5 Age vs avoidance orientation

As shown in figure (figure 25), students age 18-21 have an average factor of avoidance
orientation of 3.4. Students age 22-25 have an avoidance orientation of 3.2 and age 26-29
of 3.4. The avoidance orientation average factor drops to 2.8 for students age 30-35 and
to 2.0 for students who are older than 35 years old.

Figure 25 Age vs avoidance orientation

7.6 Nationality vs fear of failure

Nationality has some impact in fear of failure average factor. As shown in the figure (figure
26), the average factor of fear of failure for people from Asia is 3.7, 3.3 for people from
Finland. Europeans have an average factor of 3.5 of fear of failure and Africans of 3.2.
Americas have a higher average factor of fear of failure, where North America has 4.3 and
North America 4.1.

40
Figure 26 Nationality vs fear of failure

7.7 Gender comparison

In figure (figure 27), we have a comparison between two genders with all eight categories.
From a general prospective, it is noticed that there are not many differences between
males and females.

Mastery-intrinsic orientation has an average factor of 6.5 for male students and 6.4 for fe-
males.

Males have an average factor of mastery-extrinsic orientation of 5.7, while females have
5.8.

An average factor of 3.9 of performance-approach orientation is for males and 4.2 for fe-
males.

Performance-avoidance orientation has an average factor for male students of 3.7 and fe-
male of 3.6.

41
Figure 27 Gender comparison

An average factor of 3.4 of avoidance orientation is for male and 2.9 for female students.

Fear of failure has an average factor of 3.4 for males and 3.8 for females.

Academic withdrawal average factor of 3.0 both for female and male students.

School value has an average factor of 2.0 for males and 1.9 for females.

42
8 Discussion

In this chapter we are going to interpret the results by giving meaning to all the values of
the average factors. Scaling of average factors vary from zero to seven. In seven catego-
ries of the questionnaire, mastery-intrinsic orientation, mastery-extrinsic orientation, per-
formance-approach orientation, performance-avoidance orientation, avoidance orienta-
tion, fear of failure and academic withdrawal zero is low and seven is high. In school
value, zero is high and seven is low.

Questions belonging to each category, as given by Niemivirta, are available in Appendixes

1.

8.1 Academic withdrawal

Academic withdrawal is related with uncertainty that students have in school tasks.

As shown in chapter 7.1, academic withdrawal has supremacy to performance-avoidance

orientation, so students who cannot concentrate on demanding tasks during their school
tasks, are more likely to stop their studies. By taking a proper action Haaga-Helia UAS
may sway in students’ behaviours.

Students that have a high susceptibility towards academic withdrawal, they have a lower
mastery-intrinsic orientation, so they lose the motivation to learn new things.

As shown in figure (figure 21), students with a low academic withdrawal aspire to do better
in their studies and succeed. That is why academic withdrawal is not influencing mastery-
extrinsic orientation.

Academic withdrawal have a very low influence in performance-approach orientation,

which means that students who feel that they have achieved their study goal and consider
themselves capable to continue further, are not giving up easily.

8.2 Avoidance orientation

Avoidance orientation is related with students who do only the required assignments and
try to complete their studies with as less effort as they can.

43
As shown in chapter 7.2, students who want to succeed in their studies and target high
grades, are slighter likely to do only the compulsory schoolwork. So, students that have
high avoidance orientation stipulate high mastery-extrinsic orientation.

Avoidance orientation has nearly no impact in mastery-intrinsic orientation. This means

that students do not give up easily if they have ambition to learn new things and acquire
new knowledge all the time,.

BITe students who put few efforts in school tasks and try to be done with as less work as
possible have no concern about their results and others’ opinions. For them is significant
just to pass. That is why their performance-approach orientation flatten when the avoid-
ance orientation is high.

Students with a high avoidance orientation, have a high performance-avoidance orienta-

tion. Differently said, all BITe students who try to be done with the minimum work re-
quired, try to refrain from unbearable situations.

8.3 Fear of failure

As shown in chapter 7.3, students who are acquiring new knowledge during their studies,
is not influenced by fear of failure. This means that almost all the students of BITe degree
program are open to learn new things despite their worry for failing or not doing as well as
others in class.

Mastery-extrinsic orientation is higher when the fear of failure is higher. This means that
students who are agonized about exams and afraid of wrong answers want more to suc-
ceed in their studies than students that have not too much concern.

Students that have a goal to do better than others and get better results have a higher
fear of failure because they are scared if they know the right answer or if they will fail in
some tests or exams.

Performance-avoidance orientation is the category that is more pretentious by fear of fail-

ure. In other words, students who avoid situations where they feel incompetent, often
worry if they are doing worse than others. The more students get out of their comfort
zone, the less they are likely to fear different school situations.

44
8.4 School value

School value plays a very important role in students’ performance during their studies. As
shown in chapter 7.4, students are more willing to learn new things and acquire
knowledge when there is a high school value. When the school is not very useful in stu-
dents’ perspective, then they lose the interest.

There is the same situation with mastery-extrinsic orientation. Students have the goal to
succeed in their studies and get good grades when the school value is high. If the school
value is not high, then the mastery-extrinsic orientation will drop and students are not fo-
cused in doing well in their studies.

Performance-approach orientation is remaining almost the same despite school value. In

other words, students are competing between each-other, they all want to do better in
their studies and feel competent in front of others.

Performance-avoidance orientation is higher when school value is low. This means that
students try to avoid situations they feel they might be wrong or failing in front of others
when they see school useless and waste of time. So, a high school value is eliminating
those situations.

8.5 Age vs avoidance orientation

Chapter 7.5 gives some interesting facts facing age with avoidance orientation. All the stu-
dents age between 18 to 21 and 26 to 29 are the ones who are more exposed to put less
effort in school tasks and do only the required assignments. As the students get older,
they are less likely to be avoidance oriented. We can notice that students age 30 to 35 put
more effort on school works. Avoidance orientation drops drastically for students that are
more than 35 years old.

8.6 Nationality vs fear of failure

Fear of failure is an obstacle that stands between the students and their goal. This will
cause worry during classes and exams.
In chapter 7.6, we can see the visualization of fear of failure with nationality and we can
notice that students from Africa have less fear of failure during their studies than students
from other nationalities.

45
Students from North America have the highest fear of failure, followed by South America.
These students are afraid to do worse than others and they are afraid they might fail.

Europeans have a medium fear of failure.

Finnish students have a low fear of failure and this can be as a result they know well Finn-
ish educational system and the culture.

8.7 Gender comparison

The bar chart in chapter 7.7 that compares mastery-intrinsic orientation, mastery-extrinsic
orientation, performance-approach orientation, performance-avoidance orientation, aca-
demic withdrawal, fear of failure, avoidance orientation and school value between male
and female students visualises not many differences between the two genders.

We can notice that both male and female have the goal to acquire new knowledge during
their studies.

Another factor is mastery-extrinsic orientation, that is almost equally for both genders with
a high importance. This means that almost all the students have the desire and goal to do
well in school.

Performance-approach orientation has a difference between males and females. We can

conclude that female students are more demanding in grades and results than males and
they want others to think that they are competent in what they are doing.

Performance avoidance orientation is almost the same for males and females with a me-
dium average factor. This means that all BITe students are avoiding some situations they
do not give their best, but this is not a common factor.

Avoidance orientation is higher in female than male BITe students. In simpler words, fe-
male students are more likely to do only required tasks and finalize the studies with as
less effort as possible.

Fear of failure is another factor that is higher in females than in males. This means that
BITe female students are more afraid to fail in exams and worried for their progress than
males. This can come because of the degree program, but it is still a hypothesis.

46
Academic withdrawal is exactly the same for males and females with a medium average
factor. This way we understand that all BITe students have some uncertain situations, but
they do not give up easily.

School value is highly appreciated by all BITe students who took part in the survey. This is
a very important information because we understand that students find Haaga-helia UAS
as a useful polytechnic that brings value towards their studies.

8.8 Recommendations

Based on the discussions, you can find in this thesis few recommendations, but they can
change if more data is provided for analysing.

Recommendation 1:
Haaga-Helia UAS can accept more students from Africa in BITe degree program to low
down the fear of failure during studies. Students from North America have high fear of fail-
ure in this degree, so they can be suggested to attend another degree program for more
self-confidence in their studies.

Recommendation 2:
Haaga-Helia UAS can accept more female students for higher grade point average, be-
cause females are more demanding and determined to achieve better results than the
others.

Recommendation 3:
Haaga-Helia UAS can put more effort to improve school services for raising school value
in students. This way students have the goal to succeed in their studies and get good
grades.

47
9 Conclusion

The primary objective of this thesis was to make students’ performance analysis using
KNIME Analytics Platform by taking into consideration the results of the (NIEMIVIRTA,
2002) questionnaire distributed to BITe students.

We started with proof of concept by creating a KNIME project with Gjergji Make. We were
provided data from 100 students of BITe degree program taken in early 2018. To build the
project was time-efficient but coming up with future trends and predictions based on the
amount of data provided was time-consuming and yet not quite possible. For this reason,
we changed the course of this thesis to data analysis.

Analytics conducted in this thesis are based on a small amount of data which have no
time tracking on it. For this reason, we can say that making future trends predictions is not
possible but making future trend strong assumptions based on the current situation is pos-
sible.

This topic was very interesting to work with and learned many new things in data analytics
field as well as students’ performance. Surprising findings made this work interesting as
well as challenging.

To conclude, learning curve changed throughout project development from not succeed-
ing on predicting future trends due to lack of data to achieving to analyse current situation
of students’ performance.

48
References

Chapman, P., Clinton, J., Kerber, R., Khabaza, T., Reinartz, T., Shearer, C., &
Wirth, R. (2000). CRISP-DM 1.0 (Step-by-step data mining guide). CRISP-DM
Consortium. https://doi.org/10.1056/NEJMoa1108524

Copeland, M. (2016). What’s the Difference Between Artificial Intelligence,

Machine Learning, and Deep Learning? Nvidia.

Haaga-Helia UAS. (2018a). DEGREE PROGRAMME IN BUSINESS INFORMATION

TECHNOLOGY, HELSINKI. Retrieved from http://www.haaga-
helia.fi/en/education/bachelor-degree-programmes/degree-programme-business-
information-technology?userLang=en

Haaga-Helia UAS. (2018b). Organization. Retrieved from http://www.haaga-

helia.fi/en/about-haaga-helia/organization?userLang=en

Houle, P. (2016). A brief history of Machine Learning. Retrieved from

https://www.linkedin.com/pulse/brief-history-machine-learning-paul-houle/

Johnson, R. B., Onwuegbuzie, A. J., & Turner, L. A. (2007). Toward a Definition of

Mixed Methods Research. Journal of Mixed Methods Research.
https://doi.org/10.1177/1558689806298224

KNIME.COM AG. (2016). KNIME Analytics Platform.

Margaret Rouse. (2017). Data Mining. Retrieved from

https://searchsqlserver.techtarget.com/definition/data-mining

Marr, B. (2018). What Is Deep Learning AI? A Simple Guide With 8 Practical
Examples. Retrieved from
https://www.forbes.com/sites/bernardmarr/2018/10/01/what-is-deep-learning-ai-
a-simple-guide-with-8-practical-examples/#632260e18d4b

Mitchell, T. M. (2006). The Discipline of Machine Learning. Machine Learning.

https://doi.org/10.1080/026404199365326

Naik, A., & Samant, L. (2016). Correlation Review of Classification Algorithm Using
Data Mining Tool: WEKA, Rapidminer, Tanagra, Orange and Knime. In Procedia
Computer Science (Vol. 85, pp. 662–668).
https://doi.org/10.1016/j.procs.2016.05.251

Ncr, P. C., Spss, J. C., Ncr, R. K., Spss, T. K., Daimlerchrysler, T. R., Spss, C. S., &
Daimlerchrysler, R. W. (2000). Step-by-step data mining guide. SPSS Inc.
https://doi.org/10.1017/CBO9781107415324.004

NIEMIVIRTA, M. (2002). MOTIVATION AND PERFORMANCE IN CONTEXT: THE

49
INFLUENCE OF GOAL ORIENTATIONS AND INSTRUCTIONAL SETTING ON
SITUATIONAL APPRAISALS AND TASK PERFORMANCE. PSYCHOLOGIA -An
International Journal of Psychology in the Orient.
https://doi.org/10.2117/psysoc.2002.250

Orcibal, J., & Jerphagnon, L. (2018). Blaise Pascal FRENCH PHILOSOPHER AND
SCIENTIST. Retrieved from https://www.britannica.com/biography/Blaise-Pascal

Pong, S. (2017). What statistical techniques are used to perform data analysis?
Retrieved from https://www.quora.com/What-statistical-techniques-are-used-to-
perform-data-analysis

Pyle, D., & San Jose, C. (2015). An executives guide to machine learning. 2015.
https://doi.org/10.1017/CBO9781107415324.004

Roy, S., & Garg, A. (2017). Analyzing performance of students by using data
mining techniques a literature survey. 2017 4th IEEE Uttar Pradesh Section
International Conference on Electrical, Computer and Electronics (UPCON).
https://doi.org/10.1109/UPCON.2017.8251035

SAS Institute. (2018). Data Mining What it is and why it matters. Retrieved from
https://www.sas.com/en_us/insights/analytics/data-mining.html

Skillsyouneed. (2017). Simple Statistical Analysis. Retrieved from

https://www.skillsyouneed.com/num/simple-statistical-analysis.html

Sunil, R. (2016). Essentials of Machine Learning Algorithms ( with Python and R

Codes ).

Team, E. (2013). What is Educational Data Mining (EDM)? Retrieved from

http://edtechreview.in/dictionary/394-what-is-educational-data-mining

WebFinance Inc. (2018). Statistical Analysis. Retrieved from

http://www.businessdictionary.com/definition/statistical-analysis.html

50
Appendices

Appendix 1. Niemivirta questionnaire

51