Improvement of Learning Process in E-Learning Environment

Preparing to load PDF file. please wait...

0 of 0
Improvement of Learning Process in E-Learning Environment

Transcript Of Improvement of Learning Process in E-Learning Environment

International Journal of Scientific & Engineering Research, Volume 7, Issue 9, September-2016


ISSN 2229-5518

Improvement of Learning Process inELearning Environment Using Learning
Styles Recognition Approach

Farshad FarzanehKalourazi, Hew Soon Hin
Abstract—Providing adaptive tutoring system which can adapt its layout and material based on the preferred learners’ learning style is an effective way to improve the outcome of the learning process and help students to improve their performance. In this research, the web-based tutoring system offered by author collect the learner click behavior during the education and uses a machine learning algorithm in order to classify each individual learner based on their learning style. Research presented learners who participate in web-based tutoring system uses different methods in order to learn. This research applies machine learning algorithm on e-learning environment and is conducted among 50 students of Multimedia University in Malaysia who take part in web-based curriculum.

Index Terms- Learning style recognition, Felder-Silverman learning style, Neural network.



and mental behaviors that provide quite fixed indicators of the way learners understand, how

IJSER By the advent of technology everything is
transforming. One of the aspects of society that has been transforming is the ways that people learn and the method of teaching. In the recent years, there has been exponential development of webbased learning. Learners are capable to participate in many tutorial curriculums in anywhere and

they interact and respond to learning environments [9]. In the recent years, numerous learning style prototypes have been suggested [6].
A fundamental study of learning styles, exploring their credibility, reliability and effect for pedagogy, can be found in [2]. Also, investigation studies on the use of learning styles in web-based


learning provides support for the view that

Learners are identified by various learning styles

learning process can be improved through the

that they have, focusing on various kinds of

performance of resources that are consistent with a

information and the process of this information in

learner’s specific learning style [1]. For instance, in

different ways. One of the desirable features of a

many researches it has been shown that the

distance learning or eLearning system is that the

performance of learners in a Web-based learning

entire learner can learn through web-based

environment related to their self-reported learning

application in spite of their different learning

preference [5].

styles. In most of the application that was designed

There was a study that shows the learning style

for e-learning student, the user profile are similar

is not relevant to traditional learning but it is

for all learners, but each learner has different

essential in the e learning systems and the research

performance, goals, experience, abilities and needs.

shows the better effect would achieve in the

Learners have their own unique way and the

learning process where learners have assimilated

learning process is different for each learner to

and converging learning style [7].

another one. The aim of this research is to use of

machine learning algorithm in order to recognize

of the learner’s style, because the learning style

influences your level of learning success.

3 Felder-Silverman model

Learning style is recognition of treats, emotional

Many researches have used the model proposed by Felder and Silverman (1988) for education,

IJSER © 2016

International Journal of Scientific & Engineering Research, Volume 7, Issue 9, September-2016


ISSN 2229-5518

which categorizes learners based on their position

problems they face using the receive information.

on numerous measures that estimate how the

Reflective learners: these learners have a desire

learners process and perceive information:

to deeply analyze the situation and they use the

-Perception: What kind of information do the

information in their own way of thinking. They are

learners favorably perceive: including sensory

not good in team works and prefer to do things

(external) like sights, audio, physical feelings, or


intuitive (internal) such as options, visions,

-Understanding: they are usually two types of


approach towards understanding: sequentially in

Sensing learners: these learners prefer exact

constant pace, or global learners that progress in

data, facts and self-experiences. They are attracted

large jumps.

toward more procedural information and concrete,

Sequential learners: these learners usually track

practical, and are oriented towards facts. In the

a fixed way of rationale when dealing with

problem solving, these students are patiently

problems; they are interested to proceed in a direct

dealing with details and do not like to be

way. Their learning achievement would improve if

surprised. Thus, they react to the problems with a

the difficulties and problems come in a linear form.

slower pace, but they normally render a better

Global learners: global learners go through the

result [4].

problems based on their instinct and may not be

Intuitive learners: these learners are attracted

able to describe it in detail. They prefer to think

more in theoretical methods and principles. They

about the difficulties as a whole and do not

are not interested in details and do not like to solve

elaborate. These groups of learners want to get the

the difficulties and problem in mechanical. These

big picture of the problem.

types of learners are interested in innovation. They

IJSER basically do not go into the details and try to get
and instant result. They tend to be fast but on the other hand vulnerable against mistakes errors, as a result it lowers their work quality [4].
- Input: there are a number of ways by which the information is conveyed to the learners namely: visual such as images, statistical charts, sketch or verbal for example recordings and pictured videos.
Visual learners: the information in the form of pictures and graphs is more likely to be memorized and understood by this kind of learners. They prefer graphics, timelines, drafts, plans, presentations and some other visual stuff to other forms of information.
Verbal learners: it has been proved by scientists that written words are converted to verbal

The process of analyzing data was conducted, in accordance with the objectives and goals of the research and the research questions, in an attempt to implement an effective and productive analysis of the data. To find the answers to the questions, in the same manner as the studies that have been conducted before, the web-based application was used. 50 students took part in the survey and the web-based course was examined. Taking some assistance from Matlab, this research conducted a descriptive analysis in order to analyze the quantitative data that has been gathered with the help of web-based application.

equivalent and analyze them in a similar way with spoken words [3]. Therefore these learners are good at memorizing the written works as well as spoken form of information.
-Processing: generally there are two types of students first are the ones who actively take part in the group activities and the second one are those who prefer to react and are reflective.
Active learners: these learners appear more active and enthusiastic in dynamic situations. They tend to be good in team works and involve in each problem voluntarily. They want to go through any

The web-based application that collects data from student has shown in Fig 1. The web-based application gathers needed data from interaction between the students and application. The collected data will be saved in database and then are transferred to Matlab in order to process on the foundation of learning styles, the web-based was usually used with the purpose of collecting user click behavior. Furthermore, in this study, the mutual action of the learner, time, the order of some behavior and action is gathered. Then, the data is transformed into an area between 1 and 3 in

IJSER © 2016

International Journal of Scientific & Engineering Research, Volume 7, Issue 9, September-2016


ISSN 2229-5518

order to set out the power of each dimension. As

stated before the percentage of each pattern

behavior computed and used as an input to the

algorithm. The information of those students who

spent less than 10 minutes on each part of their

lesson was thrown away due to the fact according

to the teacher's timeline the studying time for each

part should be at least 10 minutes and these

students didn’t have the test's requirements.

Learning style values for each dimension were

separated into three distinctive categories. For

instance, there are global, balanced and sequential

preferences based on Felder and colleagues

suggestion. So based on their recommendations

and with regards to arranging reduction of

question for expanding reliability, these groups

Fig. 1. Magnetization as a function of applied field.

were built. The suggested thresholds by Felder are

reasonable because maximum three questions were

deleted as a result of reliability reasons.

Values which are 1 are indicated the lowest

preference in that dimension, those which are 3

indicate a preference for the other pole and finally,
IJSER the values which are 2 points to a balanced
Fig 2: sample of introduction page

The process of Training the neural network in order to achieve the accurate outputs based on the proposed inputs is a repetitive that the network is trained with several samples and the actual outputs are compared with the target and then, the weights are adjusted to produce better prediction and results. In this research, there are 50 students who participated in a web-based tutorial system and we divided them into training and test sample. After collecting data and calculating the preference of each dimension, the data are transferred to Matlab. Each of the 23 inputs has specific information about learning styles. The MFNN receives input and use activation function in order to classify learner based on learning style.
Each layer is represented as Fig 3 based on their

IJSER © 2016

International Journal of Scientific & Engineering Research, Volume 7, Issue 9, September-2016


ISSN 2229-5518


Fig 3: main structure of first layer in MFNN

Fig 5: Best Validation Performance in 10 neurons

As illustrated in Fig 5, the numbers of Hidden

Neurons are 10, but in order to decrease the rate of

Error, several tests by changing the number of

Neurons were implemented. In addition, in the

proposed model 15% of samples were selected as a

IJSER Fig 4: main structure of second layer in MFNN
The input and output of each layer are:

validation set and the number of Early-Stopping model was applied. Finally, by choosing 6 for the number of Neurons the lowest rate of error achieved (Fig 6).


Fig 6: Best Validation Performance in 6 neurons

In order to learn network 34 of the samples were chosen and 8 samples for validation and the rest of them for testing.

Radial basis function (RBF) networks generally consist of three layers: 1- Input layer, 2-hidden layer with a non-linear RBF activation function 3linear output layer.
We selected RBF neural network because of its

IJSER © 2016

International Journal of Scientific & Engineering Research, Volume 7, Issue 9, September-2016


ISSN 2229-5518

simplicity of structure and learning efficiency that is very important. The RBF neural network is able to predict in various ways. In order to test RBF neural network we have to load the input to Matlab. Based on the input, the RBF neural network learns how to classify learner. The inputs of the network are features that extracted from the web-based application. For instance, the amount of click on example, the total number of the click on self-assessment test, the sequence of clicking on exercise after learning the each part of the lecture. In addition, the RBF has 12 outputs that mean each of the dimensions is divided into 3 parts.

Tabel2: mse on various spread in RBF

Then, we used 10 samples as a test data which

have been chosen among the all of data and the

rest of them for training the network. As illustrated

in Table 3 the result is better than the previous

running but the percentage of the error is more

IJSER than MFNN neural network. Table 3: mse for 10 samples
Fig 7: Main component of RBF A class of functions form Radial functions which

could be used in any sort of model such as linear

or nonlinear and any sort of network like single-

layer or multi-layer [8] . In order to use train RBF

neural network, 35 data were chosen as samples

and 15 of the input were used as test input and the

following result were achieved. For spread=1 the

mse error was around 0.3358 that is not satisfying

and it is maybe due to the selection of test and

training data. Table 2 shows the analysis of mse

item and indicates the rate of error, based on different spread.

The examination showed that the neural network with 6 Neurons are more precise than RBF. For

instance, there are different answers in one of the

prediction (Table 4).

IJSER © 2016

International Journal of Scientific & Engineering Research, Volume 7, Issue 9, September-2016


ISSN 2229-5518

Table 4: misclassification of learning style

Based on the result, there is misclassification in

Fig 9: Target of the MFNN

IJSER one row of RBF in Table 4. In order to estimate the
efficiency of the MFNN we run the test data on MFNN. Finally, the answer of the MFNN is predicted match to target. The MFNN can be employed in order to recognize a learner’s learning style that lead to produce user interfaces based on their specific learning style in the learning environment.

The MFNN and RBF neural network make it possible to recognize automatically the preferred learning styles of learners. In other word, the behavior and interaction between learner and webbased application indicate the learning style. As a result, this learning style can help the web developer to design intelligent web-based application that is able to adapt itself to each individual learner. This result showed that 46% of students who participated are active learners that means they are interested to involve in learning process by posting in forums, doing self-

assessment test and spending time on exercise. As

mentioned in chapter 2 the active learners prefer to

try thing out by themselves. Furthermore, 32% of

them are reflective learners and 22% of them are

balanced learners. Furthermore, by the analysis of

answers, 44% of participants were visual learners

who prefer to study in a picture/video based

learning material. In addition, the result has

discovered 48% of participants were verbal learner

whose preference were text-based material and

used PDFs in order to study and finally, 8% of

them were balanced learner they study in both

Fig 8: Output of the network with 6 neurons

modes. In the Global/Sequential area, the research

shows that 50% of students were global who were

interested to get a big picture of lessons in order to

learn better and 50% of them were sequential who

prefer to study based on the structure designed by

the lecturer and follow the course step by step in a

sequential mode.

IJSER © 2016

International Journal of Scientific & Engineering Research, Volume 7, Issue 9, September-2016


ISSN 2229-5518


This study essentially uses two algorithms of


machine learning in order to classify learners based

on preferred learning style and compared them.

The methods to perform classification, as well as

[1] Budhu, M. "Interactive Web-Based

finding out the features that have the most impact

Learning Using Interactive Multimedia

on classification are important. In other words,

Simulations." Paper presented at the

which attributes are more effective in learning

International Conference on Engineering

style recognition? According to the characteristic of

Education (ICEE 2002), 2002.

each dimension and related activity, each student

[2] Coffield, F., D. Moseley, E. Hall, and K.

is put in a specific dimension. The collected data

Ecclestone. "Learning Styles and Pedagogy

are transferred to Matlab software and are trained

in Post-16 Learning: A Systematic and

to the MFNN and RBF neural network in order to

Critical Review." Learning and Skills

achieve the result. This chapter mainly focuses on

Research Centre London, 2004.

the result that gained from the classification

[3] Felder, R.M., and R. Brent.

algorithm in order to get the conclusion.

"Understanding Student Differences."

This study has investigated several features which

Journal of engineering education 94, no. 1

have an impact on the learning style recognition.

(2005): 57-72.

The result of this study can be used in order to

[4] Felder, R.M., and L.K. Silverman.

make the tutoring system more flexible based on

"Learning and Teaching Styles in

IJSER the Felder-Silverman learning style model. For
example, learners who are verbal more often used PDF and written material and therefore the webbased system usually offers the lecture in a text mode. In addition, the aim of the tutoring system established in this research is to extract learner click behavior patterns in learning environment in

Engineering Education." Engineering education 78, no. 7 (1988): 674-81. [5] Felder, R.M., and J. Spurlin. "Applications, Reliability and Validity of the Index of Learning Styles." International Journal of Engineering Education 21, no. 1 (2005): 10312.

order to develop an adaptive tutoring system. This

[6] Kolb, D.A. "Experiential Learning:

adaptive learning system can improve the learning

Experience as the Source of Learning and

process and outcome via the use of convenient way

Development." (1984).

of learning match to individual students. To attain

[7] Manochehr, N.N. "The Influence of

the goal of this research, some of important actions

Learning Styles on Learners in E-Learning

which are effective in recognizing learning style

Environments: An Empirical Study."

dimensions based on other research learning had

Computers in Higher Education Economics

been selected. This pattern behavior is based on the

Review 18, no. 1 (2006): 10-14.

Felder-Silverman theory and then using the

[8] Orr, M.J.L. "Introduction to Radial Basis

characteristics and machine learning algorithm to

Function Networks." Center for Cognitive

classify learning style. Finally, this research

Science, Scotland, UK (1996).

examines two algorithms and tries to reduce the

[9] Yau, J.Y.K., and MS Joy. "Context-Aware

error in classification and analyses the relevant

and Adaptive Learning Schedule for

pattern and the percentage of students that have

Mobile Learning." (2006).

the same dimension.

IJSER © 2016

International Journal of Scientific & Engineering Research, Volume 7, Issue 9, September-2016


ISSN 2229-5518

IJSER © 2016