Eurasian Journal of Educational Research 94 (2021) 315-332

610
Eurasian Journal of Educational Research
www.ejer.com.tr

Early Assessment of Student’s Learning Outcomes using Prediction Model under

Outcome-Based Education System

Hashim ALI1

ARTICLE INFO ABSTRACT

Article History: Purpose: Finding directions to advance in academia

Received: 23 Jul. 2020 is essential. For the appraisal of learning, assessment
Received in revised form: 12 Feb. 2021 is considerable and develops significance in
Accepted: 31 Mar. 2021 institutional higher education. This paper presents an
DOI: 10.14689/ejer.2021.94.14 intelligent real-time prediction model to assess
Keywords
student learning outcomes relevance to industry
Higher education, classification, using the Bayesian statistical inference model.
Bayesian prediction model Research Methods: A dataset of 670 students
collected from an engineering university evaluated
the proposed Bayesian-based inference model with
the conventional assessment method. The proposed
model was then evaluated based upon the prediction
accuracy and statistical kappa statistic.
Findings: This study demonstrated how students’ learning and expected success rates could
be improved during their academic careers using the presented prediction model. The
proposed methodology generated significant results with 98% accuracy and 0.94 kappa
statistic, which agreed with the traditional assessment technique.
Implications for Research and Practice: The extensive results presented which beliefs to be
an essential step towards bettering students’ academic performance and assessing the
educational program itself.

© 2021 Ani Publishing Ltd. All rights reserved

1
Pak-Austria Fachhochschule: Institute of Applied Science & Technology, PAKISTAN, e-mail:
[email protected], ORCID: 0000-0001-6116-5616
Hashim ALI / Eurasian Journal of Educational Research 94 (2021) 315-332 316

Introduction
Assessment and evaluation always have significant importance in our lives
regardless of our life stage. Academic institutions assess and evaluate students based
on the knowledge and skills they attain during their educational program. Such
evaluation involves parameters that may constitute institutions’ designed student
knowledge model, assessment methods (such as quizzes, assignments, exams, lab
work, and among others), and student learning style for each course taught to them.
However, the industry evaluates candidate profiles based on attributes/parameters
comprising specific knowledge and skills, such as problem-solving, solution design,
investigation, among others. The candidate is supposed to acquire such attributes
during their academic career. Each candidate goes through many challenges to acquire
such attributes to land a dream job. Therefore, educational institutions continuously
improve education standards to enhance student’s performance by revising curricula
associated with learning outcomes, improving faculty teaching style, incorporating
state-of-the-art laboratories infrastructure to remain in line with the industry
requirements. Each student has equal opportunities and access to academic resources.
Some students cannot grasp the delivered knowledge and skills, up to a certain level,
for many reasons, such as personal, social, and cultural reasons (Atkinson, 2000; Diaz,
2003; Georgiou et al., 2002; Gonzalez-pienda et al., 2002). The student’s learning style
depends on his learning ability, prior knowledge, and compatibility with the
faculty/instructor teaching styles (Felder and Brent, 2005; Felder and Silverman, 1988).
The conventional way of assessing students’ performance is by observing how
students learn and interact with the material either in a traditional classroom or in an
advanced online web-based environment. Some researchers (see Atkinson, 2000;
Brusilovsky and Peylo, 2003; Diaz, 2003; Georgiou et al., 2002; Helal et al., 2019; Thiele
et al., 2016) have addressed issues interlinked with an effective student knowledge
model, assessment mechanisms, and student learning styles to improve student’s
performance and success prediction.
Published work focuses on improving student performance by developing
computer-aided solutions (i.e., intelligent tutoring systems, recommended systems,
and custom-designed applications) to highlight effective student knowledge models
and related parameters. Such knowledge models assess student performance and
validate using educational data mining tools ranging from simple descriptive statistics
to advanced learning techniques. For example, the educational assessment concerning
student knowledge model started to evolve from developing Intelligent Tutoring
Systems (ITSs). ITSs are computer-based software/web applications to help students
in their learning activities. In the context of learning activity, Conati et al. (1997)
described the importance of knowledge structures embedded in the student modeling
component of ANDES, an online ITS Newtonian ‘Physics’ course. The Bayesian
Networks (BN) was used to assess knowledge and predict students’ actions during a
‘Physics’ problem-solving. Similar work was done by Gertner and VanLehn (2000). An
algorithm based on BN was developed to model student skills associated with word
processes using Desktop Associate ITS (Murray, 1999). IDEAL: another tutoring
system was used the BN technique to classify students into three categories: novice,
Hashim ALI / Eurasian Journal of Educational Research 94 (2021) 315-332 317

beginner, and intermediate (Shang et al., 2001). Feng et al. (2009) addressed different
challenges in assessing online tutoring systems using the linear regression technique.
Contrary to work published under ITS, Pardos and Heffernan (2010) proposed a
method to formulate individualization of student knowledge tracing model problems,
with low prediction errors while estimating student knowledge level.
Rapid advancement in ICT, many researchers have developed recommender
systems (Drachsler et al., 2009; Ghauth and Abdullah, 2010; Lu et al., 2018; Luo et al.,
2010; Manouselis et al., 2011; O’Mahony and Smyth, 2007; Zaiane, 2002) to help
students in both formal and informal e-learning environments. In this context, Burgos
et al. (2018) and N. Thai-Nghe et al. (2009) used recommender system techniques for
predicting student performance under ITS. The authors validated their approaches by
comparing recommender system techniques with the data mining regression methods,
such as linear logistic.
Similarly, much research was conducted over the period in estimating/predicting
students’ educational assessment. A conceptual framework was presented by Mislevy
et al. (1999) for educational assessment, based on a statistical model. Usually, statistical
methods require a large quantity of historical data for reliable predictions. A multi-
agent-based student profiling system using Fuzzy logic; Dongming Xu et al. (2002)
could store student learning activities and each student’s interaction history. Bekele
and Menzel (2005) demonstrated an application for “Mathematics” to predict student
performance using a Bayesian classifier. García et al. (2007) evaluated the Bayesian
network model to evaluate student learning style for “Artificial Intelligence” under a
Web-based education system. The model covers various aspects of student behavior
and can infer his/her learning style according to modeled behaviors. Nguyen (2016)
and Osmanbegovic & Suljic (2012) presented a comparative analysis of data mining
tools: Bayes Net, Decision Tree, and MLP over two diverse populations of students’
academic information. The results showed that such tools could predict student
performance with reliable accuracy. Neural Network techniques were used by Romero
et al. (2009) to classify students based on their Moodle usage data and the final marks
obtained in their respective courses. Nguyen Thai-Nghe et al. (2010) used support
vector machines to improve academic performance prediction by dealing with class
imbalance (i.e., the ratio between passing and failing students is usually skewed).
Kabakchieva (2013) presents preliminary student’s university performance
classification results based on personal and pre-university characteristics. Various data
mining tools, such as the J48 classifier, Bayesian classifier, k-nearest neighbor classifier,
and rule learner technique are used to assess performance. Fernandes et al. (2019)
conducted predictive data analysis of students’ academic performance in public
schools. Similar experimentation was conducted by Angiani et al. (2019) to conclude
that it is possible to predict students’ academic outcomes and intervene assisted help
case of negative performance.
The research has contributed much to strengthening knowledge models and
improving students’ learning styles; they are mostly limited to a single course either
conducted in a traditional classroom or modern environments. The significant
drawback of existing systems is that students know the evaluation outcomes at the end
Hashim ALI / Eurasian Journal of Educational Research 94 (2021) 315-332 318

of a semester or after graduation about attained industry-relevant attributes. However,

as mentioned at the start of this section, the industry evaluates candidates based on
specific attributes/parameters related to knowledge and skills that are not directly
linked with a single course. It could be the combination of multiple courses. For the
student to attain industry-relevant outcomes, accurate predictions become essential in
providing necessary assistance to students’ learning processes. An early assessment of
specific parameters; concerning industry requirements is one of the challenges, hence
to produce/anticipate important actions/recommendations, leading to success.
Consequently, there is a need to build methods that can assess attributes/parameters
directly linked with students’ skills besides knowledge.
In this connection, there is a term Outcome-based Education (OBE) system that is
relatively new compared to conventional education assessment systems. One of the
objectives of OBE is to make sure that students are acquiring sufficient skills, besides
knowledge building, and not limited to a single course or courses of the same domain.
The OBE-based system assesses a student’s performance throughout his/her academic
career within the enrolled academic program, and evaluation is performed using well-
defined attributes/parameters, known as Program Learning Outcomes (PLOs). Such
a system can assess and improve students’ academic performance and evaluate the
offered academic program’s strengths and weaknesses. To date, there is an inadequate
number of research papers that specifically address student assessment under OBE
systems, which remained under-researched. For example, the paper published by Al-
Yahya and Abdel-halim (2013) explained various procedures and arrangements
adopted by an engineering department in setting up and evaluating PLOs. The paper
explained a structured continuous assessment process and pertaining these
assessment results to strengthen the engineering program. In (C. Deneen et al., 2013;
C. C. Deneen et al., 2018), the authors discussed the difficulties and adjustments to
execute OBE. An efficient methodology for outcome-based evaluation that encourages
faculty participation while improving the assessment and reporting procedures
through powerful and meaningful visualization is introduced (Harmanani, 2017).
Moreover, the author discussed in detail the OBE system and various procedures
interlinked with it.
In the next section, brief details of PLOs are provided, which is the basis of this
research. The following research questions highlight the specific focus of this study:
1. What are the relevant industrial assessment attributes that may intervene with
student’s performance?
2. What is the educational data mining technique for acquiring values for those
identified attributes using academic records only?
3. How can we predict overall student performance based on the values in real-
time?
In summary, there is a need to devise such mechanisms that assess learning
outcomes effectively and predict relevant parameters in real-time to help students in
their pursuit of acquiring knowledge and skills. In this research study, a Bayesian-
Hashim ALI / Eurasian Journal of Educational Research 94 (2021) 315-332 319

based prediction model is presented to assess student learning outcomes under the
OBE system. An adaptive system is developed and implemented for students
struggling in their studies to increase their success probability. In comparison, the
proposed work facilitates the students and the faculty to provide assessment in real-
time. A statistical performance measure, kappa statistic, is used to evaluate the
proposed prediction system. Experimental estimates are conducted on a dataset of
students collected from a private sector national university. The predictions are based
on the outcome results, which are verified with the help of academic experts. The
proposed framework’s consequences yield a great forecast, demonstrating that the
framework model will give a critical commitment in advanced education evaluation.
The proposed research work has the following contributions:

 An educational dataset comprising all students enrolled within a single

academic program for complete program duration, verified by a team of
academic experts, is collected to assess students.

 Feature selection; relevant information was extracted for the prediction of

PLOs concerning success or failure.

 The dataset was assessed using the Bayesian inference technique to predict
students’ performance against each PLO and assess the academic program by
aggregating students’ evaluation data of the whole batch/class enrolled.

 Academic experts verified the accuracy of the prediction model.

Program Learning Outcomes w.r.t Student: Academic institutes aim to deliver

education that enriches students with the development and integration of knowledge
and skills. With the increasing number of academic institutions worldwide, education-
related procedures continuously evolve to meet new challenges of the emerging
knowledge society and cope with the industry needs. Besides institutes’ internal
efforts in maintaining the education quality, there are international accreditation
bodies under the Washington Accord, such as ABEEK, AEER, BEM, EC, ABET, ECUK,
and PEC, which evaluate the quality of education being delivered in an academic
institution. The Washington Accord is an international and multi-lateral agreement
among bodies responsible for an accrediting undergraduate engineering degree
program; currently, there are 20 full signatories. Graduate Attributes, Program
Learning Outcomes, Program Outcomes, Student Outcomes, Learning Outcomes are
the different names that describe assessment criteria (ranging from 7-12 attributes)
what students are expected to know by the time of graduation. These outcomes relate
to students’ knowledge, skills, and behaviors as they progress through the academic
program. Academic institutes adopt the ABET general students’ outcomes and slightly
modify them to suit the offered academic program, usually denoted as Program
Learning Outcomes (PLOs).
This research study is based on learning outcomes defined for an engineering
program and in line with most accreditation bodies. These PLOs consists of 12
outcomes which are 1) Engineering Knowledge 2) Problem Analysis 3)
Design/Development of Solutions 4) Investigation 5) Modern Tool Usage 6) The
Hashim ALI / Eurasian Journal of Educational Research 94 (2021) 315-332 320

Engineer and Society 7) Environment and Sustainability 8) Ethics 9) Individual and

Teamwork 10) Communication 11) Project Management and 12) Lifelong Learning. An
academic program is based on a defined curriculum, the composition of various
courses, e.g., foundation, computing, breadth, depth, and electives. Each PLO may be
assessed using single or multiple courses of the curriculum, i.e., many to many
relationships between PLOs and courses, as shown in Figure 1. The presented research
leaves discussion on course mapping/linking methodologies to the reader; however,
reasonable justification and description are provided (Al-Yahya and Abdel-halim,
2013; Harmanani, 2017).

Figure 1. Mapping of courses to PLOs

Method
Research Design
The present paper highlights the importance of predicting students' performance
based on learning outcomes (skills relevant to industry) rather than knowledge
assessment of a single course. To this aim, an adaptive system was presented based on
educational data mining techniques for early assessment of student learning. To
predict student’s performance against learning outcomes, a Bayesian-based prediction
model was designed to forecast the student’s performance in real-time. Compared
with the literature presented techniques, the proposed methodology was not limited
to students' performance in a single course. The findings suggest that the presented
system outperforms in predicting the student’s performance with an accuracy of 98%
and a kappa statistic of 0.94.
This section explains the Bayesian Network (BN) that will predict student learning
outcomes in real-time. The nodes and variables used in the network are described, and
later discussion about node relationships and parameters required for the prediction
model is explained.
Nodes: The presented network model consisted of two types of nodes: nodes to gather
evidence of student’s knowledge, which we call evidence variables, and nodes to
assess/predict student’s learning outcomes, which we call PLO variables.
Evidence Variables: The evidence variable was used to gather information about student
performance from a course. The information might be coming from different levels of
Hashim ALI / Eurasian Journal of Educational Research 94 (2021) 315-332 321

granularity within the defined student knowledge model. Granularity hierarchy

(Collins et al., 1996) provides the structure necessary to capture all course
requirements consisting of learning objectives, contents/topics, assessment methods
(e.g., quizzes, assignments, exams, presentation).
Evidence was considered a unique entity of knowledge that referred to attributes
relevant to students’ knowledge gain and associated skills. A random variable E with
Gaussian distribution was used; that is typically distributed with mean μ and variance
σ2 where μ is an unknown parameter, we wished to estimate.
Nodes Relationship and Parameters: Once the nodes of the network were described, we
had to define the relationships among them. Let us consider a PLO j; for j={1...m}, that
is assessed using a finite set of evidence E={E1...En}; consists of information gathered
from n courses. Figure 2 shows the relationship model between PLO and E1...En.

Figure 2. The general structure of BN and nodes relationship

Some parameters were required: a priori probability P(PLOj) and conditional

probabilities P(Ei|PLOj) for i={1...n} (1+n2 values) where Ei is mutually independent
against the given PLO. Positive evidence about Ei increases the probability of PLOj,
which increases the probabilities of each of the Ei. This research study is to analyze the
change in PLOs probabilities as new evidence is gathered.
Prediction Model based on Bayesian Probability: Bayesian probability is well known in
statistical computation and works on the principle of using initial belief, called prior,
and observed data generated by evidence, called likelihood, is used to approximate the
hypothesis/outcome, called posterior. Later, the initial belief is updated with the
posterior distribution. This process is known as Bayes Rule (1).

𝑃(𝐸|𝑃𝐿𝑂) · 𝑃(𝑃𝐿𝑂)
𝑃(𝑃𝐿𝑂|𝐸) = (1)
𝑃(𝐸)

Where: P(PLO) is the prior probability of hypothesis PLO being true; P(E|PLO) is
the probability that hypothesis PLO being true will result in event E; P(PLO|E) is
called the posterior probability of hypothesis PLO upon observing event E, and P(E)
is called the marginal probability which is a normalization factor.
In an academic scenario, the information: course results denoted as evidence are
coming continuously. Bayes’ rule seeks to validate or invalidate the hypothesis using
uncertain or unreliable information. We would like to apply the Bayes rule recursively,
Hashim ALI / Eurasian Journal of Educational Research 94 (2021) 315-332 322

a new data arrives and use this to reason over multiple hypotheses. We would like to
have hypotheses, PLOj={1...j}, and we seek to accumulate evidence to select the most
likely hypothesis. For multiple pieces of evidence, we could write:

𝑃(𝐸𝑖+1 |𝑃𝐿𝑂𝑗 ) · 𝑃(𝑃𝐿𝑂𝑗 )

𝑃(𝑃𝐿𝑂𝑗 |𝐸𝑖+1 , 𝑆𝑖 ) = (2)
𝑃(𝐸𝑖+1 , 𝑆𝑖 )

The problem is how to estimate P(E1...Ei) and P(E1...Ei|PLOj). To simplify the

notation, let us define S={E1...Ei } composed of i observations and E(i+1) as a recent
observation. Z-1 is a memory element to store the newly estimated posterior
probability of PLO for the next iteration. The mathematical model shown in Figure 3
is a depiction of a concept developed by E.T. Jaynes.

Figure 3. Mathematical model of posterior probability estimation of PLO

For each recent observation, E(i+1), the problem is how to estimate P(PLOj|E (i+1), Si)
using the previous evidence S. The index, i, and the accumulated evidence Si are then
updated: S(i+1)←Si∪Ei; i←i+1 and can be written as:

𝑃(𝐸𝑖+1 |𝑃𝐿𝑂𝑗 ) · 𝑃(𝑃𝐿𝑂𝑗 )

𝑃(𝑃𝐿𝑂𝑗 |𝐸𝑖+1 , 𝑆𝑖 ) = (3)
𝑃(𝐸𝑖+1 , 𝑆𝑖 )

The estimated probability predicts student success against each PLO, which
depends on the number of courses linked. The following section explains the details
of the experimental methodology using the proposed BN.

Research Sample
This section provides details of the experimental setup, including dataset
collection, data pre-processing and feature selection, integration of the proposed
prediction model, and analysis of the results. Figure 4 elaborates various steps
involved in the experimental methodology. The presented methodology is general and
can be applied to any academic program to predict student’s performance in defined
learning outcomes.
Hashim ALI / Eurasian Journal of Educational Research 94 (2021) 315-332 323

Figure 4. An overview of the experimental methodology

The student’s dataset collects all relevant information (including academic records)
gathered from the institute’s administration. Data pre-processing is an essential step
that helps in identifying missing data and outliers. Based on the experimentation, the
relevant features are extracted from the dataset to have meaningful prediction results.
Finally, the extracted features predict student’s performance in each learning outcome.
The prediction information and each PLO’s acquired probabilities for each student are
used to analyze the overall assessment. The subsequent sections explain the details of
each step.
The dataset consisting of 670 students was collected from a national university
where students were enrolled in an engineering program from 2013 to 2016 to validate
the presented model. The entire dataset was acquired under the supervision of
academic staff and verified by the examination office. Besides trivial information about
a student, the academic record contains detailed information about the students’
assessment results (i.e., assignments, quizzes, sessional exams, lab work, and others)
in each course. Each student was characterized by a set of dynamic evidence variables,
as explained earlier. The details of the dataset feature attributes are presented in Table
1. The university academic experts/policymakers set a criterion that each student must
attain at least marks in each PLO. The objective is to ensure if a student is progressing
towards success in each learning outcome. For simplicity, each PLO’s prior probability
is set as 0.5, i.e., 50%, considered as an initial belief of the institute about each student’s
performance.
Hashim ALI / Eurasian Journal of Educational Research 94 (2021) 315-332 324

Table 1
Details of Feature Attributes
Sr. No. Feature Type Value
1. Student-ID String Characters
2. Student Name String Characters
3. Semester Fall/Spring Year 2000, 2001, …
4. Assignments Integer 1, …, 10
5. Quizzes Integer 1, …, 10
6. Sessional-I Exam Integer 1, …, 20
7. Sessional-II Exam Integer 1, …, 20
8. Final Exam Integer 1, …, 40
9. Obtained Marks Integer 1, …, 100
10. CLOs Categorical: {𝐶𝐿𝑂1 , . . . , 𝐶𝐿𝑂𝑛 } Percentage
11. PLOs Categorical: {𝑃𝐿𝑂1 , . . . , 𝑃𝐿𝑂𝑚 } Percentage

Research Instruments and Procedures

Data pre-processing. An essential part while applying data prediction/learning
techniques is data pre-processing. Pre-processing is required to have meaningful
analysis and to acquire optimal results. During data cleansing, records with missing;
if a student missed any quiz/assignment or unrealistic data, unbounded data were
detected and removed. All records (student attained marks) were normalized between
0-1 to had consistency with probabilistic results. Later, the processed data were used
for further evaluation, which improved the performance of the model.
Bayesian Prediction Model. The experimentation was conducted by keeping in mind
two perspectives; one was to analyze student’s behavior for each PLO, and the second
was to evaluate the academic program by integrating all students’ performances. The
results of the prediction are evaluated using statistical metrics, i.e., accuracy and kappa
statistic. The process of the Bayesian-based prediction model is represented in Figure
5.

Figure 5. An architecture for Bayesian-based prediction model

The predictions generated by the proposed system were compared with the
outcomes labeled by the academic staff to review the correctness of the results. Two
performance metrics: accuracy and Cohen’s kappa statistic were used to determine the
Hashim ALI / Eurasian Journal of Educational Research 94 (2021) 315-332 325

proposed prediction model’s efficiency. The effectiveness of the model was evaluated
using sensitivity and specificity results. A confusion matrix was generated to evaluate
the accuracy, matching the proposed model’s results with the traditional approach’s
actual results. The experimental results showed that the proposed model, with pre-
selected prior probability (based on Gaussian distribution), estimates posterior
(predicted) probability with a high statistic value compared to the traditional
approach. It handled the missing data and outliers internally during the prediction
phase. After applying the prediction algorithm, the results concerning kappa statistics
are explained below.
The following data comes from experimentation where two independent methods,
traditional (A) and proposed (B), evaluated 840 instances. Studies A and B either said
success or failure.

 648 instances were rated by both as a success.

 178 instances were rated by both as a failure.
 Study A rated 648 instances as success and 192 instances as a failure.
 Study B rated 662 instances as success and 178 instances as a failure.
Cohen’s Kappa Statistic was calculated using the formula, Κ=(po-pe)/(1-pe) where
po is the relative observed agreement among independent studies and p e is the
theoretical probability of chance agreement. The achieved value of K is 0.94, which
means almost perfect agreement.

Data Analysis
Analysis of the model’s obtained results is an essential step to identifying students’
behavior over time. The model presented above generated the estimated probabilities
for each student against each learning outcome, which was further analyzed to predict
the estimated student success rate. Similarly, the class/batch behavior was observed
by applying the aggregation method to approximate the overall behavior of
batch/class against each learning outcome. The detailed analysis of results has been
explained in the next section.

Results
The effectiveness of the proposed model is evaluated using the proposed statistical
model. The experimental results showed that the Bayesian inference model, with
selected prior probability P(PLOi), estimated posterior P(PLOi|Ej) probability of
respective PLO using an observed data Ej. The prior probability updates with new
posterior probability would be used for newly observed data. Figure 6 shows the
trajectory of predicted probability against each course linked with attainment of PLO1.
This result’s significance is that students can have real-time visualization of predicted
probability during the academic pursuit. If the probability is less than 0.5, the academic
staff can recommend/counsel students with appropriate material to improve
student’s performance.
Hashim ALI / Eurasian Journal of Educational Research 94 (2021) 315-332 326

Figure 6. Estimated probability trajectory of PLO1

Moreover, Figure 7 shows PLO attainment trajectories of selected students against

prior probability. It is evident from the figure that an early assessment can help
students and academicians identify students who need early education assistance.

Figure 7. Real-time prediction of PLO1 by estimating the posterior against each course
(observed data) linked to attainment of PLO1

Similarly, the proposed model can track the learning outcomes of each student
during an academic session. The Bayesian predicted model’s achieved results were
evaluated with a traditional approach, as shown in Figure 8. The predicted probability
of success probability >0.5 against the traditional success rate (%age>0.5) is closely
related, highlighting the significance of the proposed methodology.
Hashim ALI / Eurasian Journal of Educational Research 94 (2021) 315-332 327

Figure 8. Comparison results of the student success rate of learning outcomes

Figure 8 presents a confusion matrix summarizing success rate discrepancies

between proposed and traditional methods. In an agreement between both methods,
the highlighted cells (in dark grey) represent true positive and true negative results;
learning outcomes assessment. Interestingly, few false negatives mean the prediction
model estimated success rate<0.5 where the traditional method has a success rate>0.5,
as highlighted in cells (light grey). The significant results are concerning false
positives, which means the probability of a positive result given an event that was not
present.

Table 2
Results of Success Rate Prediction of the Proposed Method in Comparison with the
Traditional Method
Traditional Method
Total
Rate > 0.5 Rate < 0.5
7776 0
P > 0.5 7776
Proposed (True Positive) (False Positive)
Method 168 2136
P < 0.5 2304
(False Negative) (True Negative)
Total 7944 2136 10080

It is equally important to assess the overall assessment of offered academic

program; success rate concerning all students. The academic program assessment was
analyzed using the predicted results of students. Each student’s final predicted values
against each PLO were aggregated to visualize the learning outcomes presented in
Figure 9. The figure presents an average success rate that shows all students’
cumulative success rate against the pre-defined threshold, which is 0.5 (50%).
Hashim ALI / Eurasian Journal of Educational Research 94 (2021) 315-332 328

Figure 9. Results of evaluating educational program success rate against PLOs. The average
success rate presents the cumulative success rate of all students against a pre-defined threshold

The results suggest that the overall performance of the batch/class is satisfactory
in PLOs 1,4,5,9 and marginal in PLOs 2,6,8,10,11 and critical in PLOs 3,7,12, which
requires the attention of the department. Furthermore, compared with the traditional
approach(es), it is evident from the achieved results that the proposed model can make
early predictions in case of failure. However, the system lacks in providing
recommendations to the students having lower assessment results which is the
limitation.

Discussion, Conclusion and Recommendations

The presented work here describes an adaptive prediction model’s development
to assess student learning outcomes during the pursuit of higher education. The model
utilizes pre-defined prior probabilities of each learning outcome for assessment and
estimates the posterior probabilities. The assessment and probability estimation
depends on a student’s achieved results in multiple courses and their weights, as
defined by the institution’s academic experts. This proposed work contributes to
university existing assessment methodologies by facilitating academicians in making
rapid academic decisions. Moreover, most state-of-the-art methods (Gertner and
VanLehn, 2000; Lu et al., 2018, 2018; Pardos and Heffernan, 2010) focused on
assessment using traditional approaches, while a proposed model is based on the
Bayesian inference technique. The BN model presented is neither limited to the PLOs
described here nor the courses; in fact, ‘n’ number of PLOs can be defined to assess
them from the linked courses. The system is implemented to refine educational
decision-based models’ quality by assisting academicians and students in making fast
educational decisions. The proposed methodology predicts the outcomes with a
significant accuracy of 98% than traditional approaches and literature published
results (Angiani et al., 2019; Dongming Xu et al., 2002; García et al., 2007; Nguyen,
2016; Nguyen Thai-Nghe et al., 2010) based on the gathered information. In the future,
Hashim ALI / Eurasian Journal of Educational Research 94 (2021) 315-332 329

the model can be extended by analyzing the effects of psychological parameters in

attaining high learning outcomes.

References
Al-Yahya, S. A., and Abdel-halim, M. A. (2013). A Successful Experience of ABET
Accreditation of an Electrical Engineering Program. IEEE Transactions on
Education, 56(2), 165–173. https://doi.org/10.1109/TE.2012.2206112

Angiani, G., Ferrari, A., Fornacciari, P., Mordonini, M., and Tomaiuolo, M. (2019). Real
Marks Analysis for Predicting Students’ Performance. Methodologies and
Intelligent Systems for Technology Enhanced Learning, 8th International
Conference, 804, 37–44. https://doi.org/10.1007/978-3-319-98872-6_5

Atkinson, E. S. (2000). An Investigation into the Relationship Between Teacher

Motivation and Pupil Motivation. Educational Psychology, 20(1), 45–57.
https://doi.org/10.1080/014434100110371

Bekele, R., and Menzel, W. (2005). A Bayesian Approach to Predict Performance of a

Student (BAPPS): A Case with Ethiopian Students. Algorithms, 22, 24.
https://nats-www.informatik.uni-
hamburg.de/~menzel/papers/aia2005.pdf

Brusilovsky, P., and Peylo, C. (2003). Adaptive and Intelligent Web-based Educational
Systems. International Journal of Artificial Intelligence in Education, 13(2–4), 159–
172.

Burgos, C., Campanario, M. L., Peña, D. de la, Lara, J. A., Lizcano, D., and Martínez,
M. A. (2018). Data Mining for Modeling Students’ Performance: A Tutoring
Action Plan to Prevent Academic Dropout. Computers & Electrical Engineering,
66, 541–556. https://doi.org/10.1016/j.compeleceng.2017.03.005

Collins, J. A., Greer, J. E., and Huang, S. X. (1996). Adaptive Assessment Using
Granularity Hierarchies and Bayesian Nets. Intelligent Tutoring Systems, 569–
577. https://doi.org/10.1007/3-540-61327-7_156

Conati, C., Gertner, A. S., VanLehn, K., and Druzdzel, M. J. (1997). On-Line Student
Modeling for Coached Problem Solving Using Bayesian Networks. User
Modeling, 381, 231–242. https://doi.org/10.1007/978-3-7091-2670-7_24

Deneen, C., Brown, G. T. L., Bond, T. G., and Shroff, R. (2013). Understanding
Outcome-Based Education Changes in Teacher Education: Evaluation of a
New Instrument with Preliminary Findings. Asia-Pacific Journal of Teacher
Education, 41(4), 441–456. https://doi.org/10.1080/1359866X.2013.787392

Deneen, C. C., Brown, G. T. L., and Carless, D. (2018). Students’ Conceptions of

Eportfolios as Assessment and Technology. Innovations in Education and
Hashim ALI / Eurasian Journal of Educational Research 94 (2021) 315-332 330

Teaching International, 55(4), 487–496.

https://doi.org/10.1080/14703297.2017.1281752

Diaz, A. L. (2003). Personal, Family, and Academic Factors Affecting Low

Achievement in Secondary School. Electronic Journal of Research in Educational
Psychology and Psychopadology, 1(1), 43–66.

Dongming Xu, Huaiqing Wang, and Kaile Su. (2002). Intelligent Student Profiling with
Fuzzy Models. Proceedings of the 35th Annual Hawaii International
Conference on System Sciences, 8.
https://doi.org/10.1109/HICSS.2002.994005

Drachsler, H., Hummel, H. G. K., and Koper, R. (2009). Identifying the Goal, User
Model and Conditions of Recommender Systems for Formal and Informal
Learning. Journal of Digital Information, 10(2).
https://journals.tdl.org/jodi/index.php/jodi/article/view/442

Felder, R. M., and Brent, R. (2005). Understanding Student Differences. Journal of

Engineering Education, 94(1), 57–72. https://doi.org/10.1002/j.2168-
9830.2005.tb00829.x

Felder, R. M., and Silverman, L. K. (1988). Learning and Teaching Styles in Engineering
Education. Engineering Education, 78(7), 674–681.

Feng, M., Heffernan, N., and Koedinger, K. (2009). Addressing the Assessment
Challenge with an Online System That Tutors as It Assesses. User Modeling
and User-Adapted Interaction, 19(3), 243–266. https://doi.org/10.1007/s11257-
009-9063-7

Fernandes, E., Holanda, M., Victorino, M., Borges, V., Carvalho, R., and Erven, G. V.
(2019). Educational Data Mining: Predictive Analysis of Academic
Performance of Public School Students in the Capital of Brazil. Journal of
Business Research, 94, 335–343. https://doi.org/10.1016/j.jbusres.2018.02.012

García, P., Amandi, A., Schiaffino, S., and Campo, M. (2007). Evaluating Bayesian
Networks’ Precision for Detecting Students’ Learning Styles. Computers &
Education, 49(3), 794–808. https://doi.org/10.1016/j.compedu.2005.11.017

Georgiou, S. N., Christou, C., Stavrinides, P., and Panaoura, G. (2002). Teacher
Attributions of Student Failure and Teacher Behavior Toward the Failing
Student. Psychology in the Schools, 39(5), 583–595.
https://doi.org/10.1002/pits.10049

Gertner, A. S., and VanLehn, K. (2000). ANDES: A Coached Problem Solving

Environment for Physics. International Conference on Intelligent Tutoring
Systems, 133–142. https://doi.org/10.1007/3-540-45108-0_17
Hashim ALI / Eurasian Journal of Educational Research 94 (2021) 315-332 331

Ghauth, K. I., and Abdullah, N. A. (2010). Learning Materials Recommendation Using

Good Learners’ Ratings and Content-Based Filtering. Educational Technology
Research and Development, 58(6), 711–727. https://doi.org/10.1007/s11423-
010-9155-4

Gonzalez-pienda, J. A., Nunez, J. C., Gonzalez-pumariega, S., Alvarez, L., Roces, C.,
and Garcia, M. (2002). A Structural Equation Model of Parental Involvement,
Motivational and Aptitudinal Characteristics, and Academic Achievement.
The Journal of Experimental Education, 70(3), 257–287.
https://doi.org/10.1080/00220970209599509

Harmanani, H. M. (2017). An Outcome-Based Assessment Process for Accrediting

Computing Programmes. European Journal of Engineering Education, 42(6),
844–859. https://doi.org/10.1080/03043797.2016.1226781

Helal, S., Li, J., Liu, L., Ebrahimie, E., Dawson, S., and Murray, D. J. (2019). Identifying
Key Factors of Student Academic Performance by Subgroup Discovery.
International Journal of Data Science and Analytics, 7(3), 227–245.
https://doi.org/10.1007/s41060-018-0141-y

Kabakchieva, D. (2013). Predicting Student Performance by Using Data Mining

Methods for Classification. Cybernetics and Information Technologies, 13(1), 61–
72. https://doi.org/10.2478/cait-2013-0006

Lu, O. H. T., Huang, A. Y. Q., Huang, J. C. H., Lin, A. J. Q., Ogata, H., and Yang, S. J.
H. (2018). Applying Learning Analytics for the Early Prediction of Students’
Academic Performance in Blended Learning. Journal of Educational Technology
& Society, 21(2), 220–232.

Luo, J., Dong, F., Cao, J., and Song, A. (2010). A Context-Aware Personalized Resource
Recommendation for Pervasive Learning. Cluster Computing, 13(2), 213–239.
https://doi.org/10.1007/s10586-009-0113-z

Manouselis, N., Drachsler, H., Vuorikari, R., Hummel, H., and Koper, R. (2011).
Recommender Systems in Technology Enhanced Learning. In Recommender
systems handbook (pp. 387–415). Springer US. https://doi.org/10.1007/978-0-
387-85820-3_12

Mislevy, R. J., Almond, R. G., Yan, D., and Steinberg, L. S. (1999). Bayes Nets in
Educational Assessment: Where the Numbers Come From. Proceedings of
the Fifteenth Conference on Uncertainty in Artificial Intelligence, 437–446.
http://dl.acm.org/citation.cfm?id=2073796.2073846

Murray, W. (1999). An Easily Implemented, Linear-Time Algorithm for Bayesian

Student Modeling in Multi-Level Trees. Proc. of 9th World Conference of
Artificial Intelligence and Education AIED, 99, 413–420.
Hashim ALI / Eurasian Journal of Educational Research 94 (2021) 315-332 332

Nguyen, L. T. (2016). Confidence-Based Assessment for Learning in ePortfolio

Environment. 2016 5th IIAI International Congress on Advanced Applied
Informatics (IIAI-AAI), 328–331. https://doi.org/10.1109/IIAI-AAI.2016.213

O’Mahony, M. P., and Smyth, B. (2007). A Recommender System for On-line Course
Enrolment: An Initial Study. Proceedings of the 2007 ACM Conference on
Recommender Systems, 133–136. https://doi.org/10.1145/1297231.1297254

Osmanbegovic, E., and Suljic, M. (2012). Data Mining Approach for Predicting Student
Performance. Economic Review: Journal of Economics and Business, 10(1), 3–12.

Pardos, Z. A., and Heffernan, N. T. (2010). Modeling Individualization in a Bayesian

Networks Implementation of Knowledge Tracing. UMAP’10 Proceedings of
the 18th International Conference on User Modeling, Adaptation, and
Personalization, UMAP 2010, 255–266. https://doi.org/10.1007/978-3-642-
13470-8_24

Romero, C., González, P., Ventura, S., del Jesus, M. J., and Herrera, F. (2009).
Evolutionary Algorithms for Subgroup Discovery in e-Learning: A Practical
Application Using Moodle Data. Expert Systems with Applications, 36(2, Part
1), 1632–1644. https://doi.org/10.1016/j.eswa.2007.11.026

Shang, Y., Shi, H., and Chen, S.-S. (2001). An Intelligent Distributed Environment for
Active Learning. Journal on Educational Resources in Computing (JERIC), 1(2es),
4.

Thai-Nghe, N., Busche, A., and Schmidt-Thieme, L. (2009). Improving Academic

Performance Prediction by Dealing with Class Imbalance. 2009 Ninth
International Conference on Intelligent Systems Design and Applications,
878–883. https://doi.org/10.1109/ISDA.2009.15

Thai-Nghe, Nguyen, Drumond, L., Krohn-Grimberghe, A., and Schmidt-Thieme, L.

(2010). Recommender System for Predicting Student Performance. Procedia
Computer Science, 1(2), 2811–2819.
https://doi.org/10.1016/j.procs.2010.08.006

Thiele, T., Singleton, A., Pope, D., and Stanistreet, D. (2016). Predicting Students’
Academic Performance Based on School and Socio-Demographic
Characteristics. Studies in Higher Education, 41(8), 1424–1446.
https://doi.org/10.1080/03075079.2014.974528

Zaiane, O. R. (2002). Building a Recommender Agent for e-Learning Systems.

International Conference on Computers in Education, 2002. Proceedings., 55–
59 vol.1. https://doi.org/10.1109/CIE.2002.1185862