Applying bayesian neural network to evaluate the influence o f specialized mini projects on final perform ance o f engineering students a case study

■ MATHEMATICS AND COMPUTER SCIENCEI COMPUTATIONAL SCIENCE, PHYSICAL SCIENCES I ENGINEERING DOI: 10.31276/VJSTE.644.10-15Applying Bayesian neural network to evaluate the influence o f

■ MATHEMATICS AND COMPUTER SCIENCEI COMPUTATIONAL SCIENCE, PHYSICAL SCIENCES I ENGINEERING DOI: 10.31276/VJSTE.64(4).10-15

Applying Bayesian neural network to evaluate

the influence o f specialized mini projects on final

perform ance o f engineering students: A case study

‘University o f Sciences, Vietnam National University, Hanoi 2Faculty o f Building and Industrial Construction, Hanoi University o f Civil Engineering

Received 10 June 2022; accepted 8 September 2022

A bstract:

In this article, deep learning probabilistic models are applied to a case study on evaluating the influence of specialized mini projects (SMPs) on the performance o f engineering students on their final year project (FYP) and cumulative grade point average (CGPA) This approach also creates a basis to predict the final performance

of undergraduate students based on their SMP scores, which is a vital characteristic o f engineering training The study is conducted in two steps: (i) establishing a database by collecting 2890 SMP and FYP scores and the associated CGPA o f a group o f engineering students that graduated in 2022 in Hanoi; and (ii) engineering two deep learning probabilistic models based on Bayesian neural networks (BNNs) with the corresponding architectures o f 8/16/16/1 and 9/16/16/1 for FYP and CGPA, respectively The significance o f this study is that the proposed probabilistic models are capable o f (i) providing reasonable analysis results such as the feature importance score o f individual SMPs as well as an estimated FYP and CGPA; and (ii) predicting relatively close estimations with mean relative errors from 6.8% to 12.1% Based on the obtained results, academic activities to support student progress can be proposed for engineering universities.

Introduction

N ow adays, universities are capable o f collecting data w ith

reference to their students in electronic format A s a result,

there is an urgent need to effectively transform large volumes

o f data into know ledge to im prove the quality o f m anagerial

decisions and to predict academic perform ance o f students

at an early stage As a part o f artificial intelligence (AI)

techniques recently adopted in a w ide variety o f hum an life

applications [1,2], various m achine learning (M L) approaches

have been increasingly applied to analyse educational data,

such as student scores, to concentrate academ ic assistance

on students as w ell as to im prove the university training

program s ML is an especially appealing alternative in the

field o f engineer training and education as it is difficult or

unfeasible to develop conventional algorithm s to perform

required tasks [3 ,4 ],

S.S Abu-Naser, et al (2015) [4] developed an artifical

neural netw ork (ANN) m odel for predicting student

perform ance at the Faculty o f Engineering and Inform ation

Technology, A l-A zhar University, based on the registration records o f 1407 students using a feed forw ard back propagation algorithm for training The m odel w as tested

w ith an overall result o f 84.6% E.Y Obsie, et al (2018) [5] developed a neural netw ork m odel for predicting student cum ulative grade point averages for the 8th sem ester (CGPA8) and designed an application based on the predictive models The real dataset em ployed in the study was gathered from

134 students at the H aw assa U niversity School o f Com puter Science that graduated in 2015, 2016, and 2017 It is shown that the student progress perform ance, w hich is m easured

by CGPA8, can be predicted using scores from their first-, second-, and third-year courses Z Iqbal, et al (2017) [6] utilized collaborative filtering (CF), m atrix factorization (M F), and restricted Boltzm ann m achine (RBM ) techniques

to system atically analyse real-w orld data collected from 225 undergraduate students enrolled in the Electrical Engineering program at the Inform ation Technology U niversity (ITU) from w hich the academ ic perform ance o f the ITU students was evaluated It was shown that the RBM technique was

'Corresponding author: Email: thangnt2@huce.edu.vn

I 1 ■.!rral ntScitnct.

" .> jt: t ncintt-rin"

D E C E M B E R 2 0 2 2 • V O L U M E 6 4 N U M B E R 4

MATHEMATICS AND COMPUTER SCIENCE | COMPUTATIONAL SCIENCE, PHYSICAL SCIENCES I ENGINEERING m

better than the other techniques in predicting student

perform ance in the particular course S.D.A Bujang, et

al (2021) [7] introduced a com prehensive analysis o f

m achine learning techniques to predict final student grades

in first sem ester courses by im proving the perform ance o f

predictive accuracy The perform ance accuracy o f six well-

know n m achine learning techniques, namely, decision tree

(J48), support vector m achine (SVM ), naive bayes (NB),

K-nearest neighbour (K-NN), logistic regression (LR), and

random forest (RF) using 1282 student course grades were

presented and follow ed by a m ulticlass prediction m odel to

reduce the over-fitting and m isclassification results caused by

im balanced m ulti-classification using the Synthetic M inority

O versam pling Technique (SM OTE) w ith a tw o feature

selection m ethod It w as shown that the proposed model

integrated w ith RF had significant im provem ent w ith the

highest f-m easure o f 99.5% [7],

It is w orth m entioning that m ost o f the aforem entioned

M L approaches w ere conducted in a determ inistic manner

Hence, there is a need to develop a probabilistic m odel that

is capable o f providing w ell-predicted results as well as

estim ating the confidence o f the results through associated

intervals Such a result is m ore relevant for experim ental data

on student exam scores rather than a single point estim ation

because, even w ith the same student, scattered results can be

obtained from different series o f experiments

In training programs at engineering universities,

specialized mini projects (SM Ps) play an im portant role as

they progressively provide knowledge as well as accumulate

conceiving, designing, implementing, and operating skills

necessary for their FYP, w hich is an integrated topic to solve

a practical problem o f the particular field o f engineer training

This article applies a m achine learning approach to predict FYP

and final CGPA results from those SMPs based on w hich the

influence o f SMPs on the FYP and CGPA can be evaluated in

a data-driven manner A case study is conducted by collecting

2890 datapoints in the form o f score results from eight SMPs,

one FYP, and the CGPA o f a group o f 289 engineering students

that graduated in 2022 in Hanoi Then, two deep learning

probabilistic models based on BNNs are established for FYP

and CGPA predictions It is shown from the obtained results

that the proposed approach is a practical tool providing quick

and reasonable analysis results such as the feature importance

score o f an individual SMP and the estim ated FYP and CGPA

results Furthermore, a relatively close estim ation can be

captured from the BNN model for CGPA, providing useful

information for academic management

Stochastic model using BNNs

As pointed out by various authors, a m ajor obstacle to the

data-driven m ethod is the scarcity o f relevant data, and this

problem becom es accentuated w hen studying the obtained

scores o f various individual students [8-12], Even w ith data

in hand, there exist unavoidable deviations betw een them [13-14], Thus, this study proposes to engineer a probabilistic

m achine learning m odel on the basis o f BNNs rather than determ inistic ones as done in the review ed works The advantage o f such a probabilistic m odel is that it is capable

o f predicting quantities o f interest such as the FYP score or CGPA, as well as estim ating the am ount o f uncertainty that

is associated w ith the prediction values It is evident that the

m ore data available, the m ore accurate the m odel, and vice versa In summary, the key contribution o f this article is to propose a probabilistic M L m odel to predict the FYP score and CGPA results from the given scores o f SMPs so that the effect o f an individual SMP as w ell as FYP on the final perform ance o f students can be evaluated

We begin by briefly review ing the A N N to set up

m athem atical sym bols and terminology G iven a dataset

w ith n being the num ber o f features Herein, each feature is

an input related to SM P and FYP results It is desirable to develop a non-linear m apping from X to Y, i.e., Y=fiX.) A standard architecture o f A N N consists o f an input layer, an output layer, and one or m any hidden layers w ith the total num ber o f layers o f the A N N being L. Each layer consists

o f various neurons that are fully connected w ith all neurons

in neighboring layers M athem atically, a neuron j at layer / could be described by a linear transform ation plus a non­ linear activation function, as follows:

where x} is output o f neuron j at layer /; is output o f neuron k at the previous layer /-1; w ! k is a w eight assigned

to the connection betw een the form er and the latter; Nj j is the total num ber o f neurons o f layer /- l; fa/-1 is a real value

to be determ ined, also know n as bias; and h is an activation function Here, the sigm oid activation function is used to squeeze values into the range (0,1) M oreover, the function is continuous and differentiable everywhere, thus rendering the training process o f the neural netw ork via a gradient descent- based algorithm faster and m ore effective

By setting W as the m atrix o f weights corresponding to layer /, w ith /= 1 , ,L , the A N N could be described by the following equation derived from the description o f neural netw orks in [15]:

w here ?t is a prediction o f F, and f t w ith l= \, ,L denotes transform ation operations at layer / in the ANN The netw ork will be iteratively trained to determ ine the optim al values o f

Wl that m inim ize the discrepancy betw een V) and Y

BNN is a probabilistic deep learning m odel that combines the high prediction perform ance o f A N N w ith the ability to

D E C E M B E R 2 0 2 2 • V O L U M E 6 4 N U M B E R 4 Vietnam Journal of Science.

Technology and Engineering 11

■ MATHEMATICS AND COMPUTER SCIENCE COMPUTATIONAL SCIENCE, PHYSICAL SCIENCES ENGINEERING

estimate uncertainty o f the Bayes theory [16] In the authors’

opinion, the model is especially suitable for working with

not-so-abundant collected data owing to two reasons: (i) in

practice, similar series o f experiments w ith identical input

parameters still provide different results due to unavoidable

uncertainty; and (ii) fitting an ANN with many parameters to a

limited database m ay cause the over-fitting problem, i.e., ANN

is likely to yield low-accuracy results on new data despite being

well trained In other words, it is necessary to not only perform

prediction o f FYP and CGPA results but also to estimate how

much confidence we have about the prediction results

For this purpose, rather than assigning the deterministic

values for weight W o f the neural network, BNN uses a

Gaussian probability distribution for W as below:

W = p + cr x e withe~N{0,l) ( 3 )

where // and a denote the m ean and standard deviation

m atrices o f W and e is the noise drawn from a zero-m ean

unity-variance norm al distribution Then, p, a are param eters

to determ ine through the learning process

N ote that the output o f BNN is a probability distribution,

thus a specialized loss function L is required to measure the

m odel’s perform ance The adopted m etric is the Kullback-

Leibler divergence (KL) w hose form ula is:

Next, the optimal values o f p" and a" are the solutions o f

the following m inim ization problem:

p ' o * = a rg m in K L ( q ( W \p , a ) \\p ( W \D ') ) ( 5 )

n,0

Via the B ayes’ rule, p(W\D) can be calculated as below:

p {W \D ) = p (D \W )p (W )

Substituting Eq (6) into Eq (4), the loss function L is

rewritten as follows:

L loSP(D) ~ 'ogPW - 'ogpfDllV) (7)

This loss function can be approxim ated from observed

discrete data as follows:

L = — ^ [ l o g q ( W ild er) - lo g p ( W 0 - logp(D |W D ] ( 8 )

s i= l

where A is the total num ber o f samples

N ext, the gradients o f the loss function w ith respect to p

and a are derived by:

d L (W \n ,a ) t d L (W \n ,a )

d L (W \p ,a ) dL (W \p , a )

A a = — d W — x e + - 3 - do

Finally, p and a are updated using a small learning rate a

as follows:

a <- a — a x Act; p <- p — a x Ap. (10)

Case study - Database on score results of SMPs, FYP and CGPA

In the postgraduate training program o f civil engineers,

a student is required to pass all the subjects including eight specialized mini projects before being qualified to conduct his/her final year project The SM Ps consist of: Design

o f Architecture (DoA), D esign o f Foundation (DoF),

M echanism o f Reinforced Concrete Structures (M oRCS); Design o f Reinforced Concrete Buildings (DoRCB); Design o f Structural Steel Buildings (DoSSB); Construction Technology 1 (C T l); Construction Technology 2 (CT2); and Construction M anagem ent (CM) that will be num bered from

SM P.l to SMP.8, respectively Each individual SMP provides students w ith the corresponding know ledge and professional skill that w ill be integrated in his/her FYP to design and build

a civil/industrial building in real situation It is notew orthy that the SM Ps’ and F Y P ’s exams are all conducted in the form o f oral defence As a result, all the aforem entioned SMPs significantly influence the FYP, which together with all theoretical subjects and SM Ps contributes to the CGPA (Fig 1)

Fig 1 Projects in training program o f civil engineers in HUCE.

In this research, a dataset o f 2890 scores o f eight SMPs, one FYP, and the CGPA is collected from 289 civil engineers

w ho graduated from Hanoi U niversity o f Civil Engineering (HUCE) in 2022 Figs 2-5 display the histogram s o f all the score results o f their SMPs, FYP, and CGPA on the 4-point scale, showing clear visualization o f the range o f values as well as their distributions

140 120

100

'C 80

a a g* 60 ill

40

20 0 1.0 1.5 2.0 2.5 3.0 3.5 4.0

S c o re r e s u lts S c o re r e s u lts

Fig 2 Score histogram s of S M P.1-D oA and SWIP.2-DoF.

Vietnam Journal o f Science, DECEMBER 2022 VOLUME 64 NUMBER 4

MATHEMATICS AND COMPUTER SCIENCE COMPUTATIONAL SCIENCE, PHYSICAL SCIENCES ENGINEERING U

Fig 3 Score histogram s of SMP.3-MoRCS, SMP 4 -D 0 RCB and

SMP 5 -D 0 SSB.

140

120

100

S '80

V 60

to.

40

20

Fig 4 Score histogram s of SNIP.6-CT1, SMP.7-CT2 and SMP.8-CM.

Score results

Fig 5 Histogram s of FYP and CGPA results.

It can be seen in Fig 2 that the score results o f SMP.I-DoA

o f the investigated 289 students were quite high, among which

74 students obtained 4.0, whereas the biggest group was 3.0

with 111 students There were 67 students that earned a 3.5 and

the number o f students receiving scores o f 2.5, 2.0, and 1.5

were 11,21, and 5, respectively It is noted that although this is

the first SMP o f engineering students in the university program,

there is not much calculation in this task of architectural design

The remaining seven SMPs are all critical to students as they

are curtailed to train them on the design as well as construction

o f buildings It can be observed in the distributions o f these

SMP scores that the group o f 3.0 is almost dominant, except the

cases o f SMP.2-DoF and SMP.5-DoSSB, o f which the dominant

group was 2.0 (Figs 2 and 3) It is shown in Figs 3 and 4 that

the distribution o f the main SMPs in the program were similar

to each other Meanwhile, it can be observed in Fig 5 that the

distribution o f FYP results was quite standard, whereas there

is a regression trend o f the number o f students with the higher

scores in the CGPA results

Analysis results on FYP and CGPA using BNN model

In this study, the adopted architecture o f the BNN for FYP

scores is 8/16/16/1 The model consists o f an input layer with 8

neurons, 2 hidden layers with 16 neurons, and an output layer

with one neuron corresponding to the FYP result, as graphically illustrated in Fig 6A The neurons in the input layer correspond

to the score results o f SMP.l to SMP.8 Since the data size is moderate, it is reasonable to avoid using too deep architectures

o f many hidden layers as well as wide layers with a significant number o f neurons, which may lead to a pronounced increment

o f parameters to determine Besides, the number o f neurons for the hidden layer is set to 16 since it should be a power o f 2 to be convenient for the memory o f the computer For the proposed BNN model, each neuron has two parameters to be determined, characterizing the probability distribution o f its weight At the beginning o f learning, they are initialized as a normal distribution with zero mean and unity standard deviation For prediction o f CGPA, the corresponding BNN architecture is 9/16/16/1 in which the FYP score in the dataset is combined with the SMPs as the 9th element o f the input layer (Fig 6B)

(A) BNN model for FYP score (B) BNN model for CGPA (C) BNN protocol

Fig 6 Graphical representation of BNN whose weights are characterized by probability distributions.

U pdating the m o d el’s w eights as described in Eq.(10) is based on the A dam O ptim ization A lgorithm belonging to the first-order gradient descent optim ization family, w hich gradually adapts the m o d el’s w eights by a sm all am ount after each iteration to reduce the loss function The num ber

o f updates is controlled through a hyper-param eter o f 0.001 This value can also be referred to as the learning rate that

w as determ ined via a prelim inary test to ensure the learning process is convergent w ithin a reasonable learning time It is noted that a sm all learning rate will unnecessarily increase learning tim e, w hereas a large value could lead to prem ature results On the other hand, pre-processing standardization

is adopted to obviate the scale difference issue betw een different features w ith different physical m eanings B y utilizing the aid o f the deep learning library Pytorch for building the overall fram ew ork, the deep probabilistic library Pyro [17] for establishing the B N N -based data- driven m odel, Pandas for data m anagem ent, scikit-leam library [18] for data standardization, and M atplotlib for data visualization, the im plem entation o f the proposed data- driven fram ew ork can be realized in this study

In this study, the investigated database is split into three non-overlapping datasets, nam ely, the training, validation, and testing sub-sets w ith a ratio o f 60:20:20, corresponding

to data from 173, 58, and 58 students

DECEMBER 2022 VOLUME 64 NUMBER 4 Vietnam Journal o f Science,

Technology and Engineering 13

■ MATHEMATICS AND COMPUTER SCIENCE COMPUTATIONAL SCIENCE, PHYSICAL SCIENCES ENGINEERING

A fter being built, the probabilistic m odels w ere trained

w ith the training database m entioned in the previous section

Figs 7A and 7B depict the learning curves o f the B N N

m odels o f FY P and CGPA results, respectively, show ing

how loss functions evolve versus the num ber o f training

iterations (epochs) on both training and validation datasets

In the follow ing paragraphs, the corresponding results o f

the BNN m odel for CGPA will be given in parentheses

The KL loss function o f FYP (CGPA) quickly dropped for

epochs from 0 to 500, before gradually decreasing to values

around 0.035 and 0.02 (0.03 and 0.025) on the training and

validation datasets, respectively A fter that, a steady trend

is observed, i.e., no clear im provem ent is obtained, until

the num ber o f epochs reached 2000 For epochs 1300 to

1400, there is less fluctuation in loss function than the other

intervals H ence, an iteration value o f 1300 is selected as the

final configuration o f the B N N m odel

0.175 0.150 0.125

8 o.ioo

^ 0.075 0.050 0.025

N um ber of iterations Num ber of iterations

(A) BNN model for FYP (B) BNN model for CGPA

Fig 7 Evolution of the KL loss function against num ber of

iterations on training and validation datasets fo r (A) FYP and (B)

CGPA.

The effect o f each SM P on FY P and CGPA results can

be investigated via a feature im portance study Im portance

scores will be assigned for all SM Ps, a high score m eans that

the corresponding SM P has a significant im pact on the FY P/

CGPA and a low score m eans there is less impact The SM Ps

are ranked based on their im portance score Such results

provide understanding about the correlation betw een the

projects and helps students optim ize their learning strategy

to achieve desired final scores The perm utation feature

im portance m ethod is used along w ith the proposed data-

driven m odel H erein, a feature refers to an SM P score The

sim ple, yet effective, core idea o f this m ethod is to perm ute

the values o f features and evaluate the change in prediction

errors A perm uted feature m eans that original values o f this

feature are shuffled am ong data sam ples w hile other features

rem ain unchanged I f a perm uted feature incurs large errors,

this feature is im portant and contributes significantly to

the prediction results H ence, all features will be perm uted

one by one, and the respective error will be calculated

for each case N ext, these errors are sorted in descending

order, and im portance scores are derived Since the BNN

is a probabilistic m odel, the feature im portance results

obtained w ith BNN are random variables Thus, one repeats

the feature im portance calculations 100 tim es and then

derives their statistical characteristics such as m ean, min, and m ax values The influence o f SM Ps on a student’s final perform ance on their FY P and CGPA in term s o f im portance score are show n in Fig 8A and Fig 8B, respectively

Feature im portance on FYP Feature importance on CGPA

0.05 0.10 0.15 0.20 0.25 0.30

Im portance score (A) BNN model for FYP

0.2 0.3 0.4 0.5 0.6

Im portance score (B) BNN model for CGPA

Fig 8 Feature im portance graph of S M P s’ influence on (A) FYP and (B) CGPA.

It can be observed from Fig 8A that, com pared to the rem aining SM Ps, the m ini projects o f DoSSB and M oRCS have m ore pronounced influence on FYP results It is noted that in the training program the SM P and FY P have 1 and

10 credits, respectively (each credit equals to 15 hours o f lecturing) It can be seen in Fig 8B that the influence o f the FYP on CGPA, w hich is counted from a total o f 168 credits,

is m ost significant w hile M oR C S holds the second position The large gap betw een the feature im portance scores o f FYP and M oR C S is reasonable due to their credit relative ratio o f 10:1 Fig 8 also proves that reinforced concrete and steel structures are the leading applications w ith significant distributions currently in the field o f construction

N ext, the final perform ance o f the trained m odel is evaluated on the test dataset and the prediction results o f

FY P and CGPA are dem onstrated in Fig 9

(A ) BNN m odel fo r F Y P (B) BNN m odel f o r C G P A

Fig 9 Prediction results o f (A) FYP scores and (B) CGPA.

It is show n in Fig 9 th a t fo r each data p o in t (sh o w n

in the d o t sy m b o l), its X -c o o rd in a te d e n o tes tru e F Y P sco res fro m the d atab ase w h ile the Y -co o rd in ate is a

v a lu e p re d ic te d b y the m o d el Ideally, a p e rfe c t m odel

w ill p ro v id e the sam e resu lts as th o se fro m the d atab ase,

as h ig h lig h te d by the re d 4 5 -d eg ree line W h ile the B N N

m o d el fo r F Y P is n o t as clo se as e x p e c te d in F ig 9A , the

Vietnam Journal of Science,

Tethnologt and Engineering

DECEMBER 2022 VOLUME 64 NUMBER 4

MATHEMATICS AND COMPUTER SCIENCE I COMPUTATIONAL SCIENCE, PHYSICAL SCIENCES I ENGINEERING ■

p re d ic te d p o in ts o f C G P A lie re la tiv e ly clo se to the ideal

lin e in Fig 9B S pecifically, th e m e a n re la tiv e erro rs o f

p re d ic te d F Y P an d C G P A re su lts are 12.1% a n d 6.8% ,

resp ectiv ely T h ese re su lts q u a lita tiv e ly co n firm the

v ia b ility o f the p ro p o se d B N N m o d e l fo r C G PA

Conclusions

T h is stu d y a p p lie d a d a ta -d riv e n m e th o d fo r a ssessin g

th e F Y P sco re a n d C G P A b a se d o n th e re su lts o f eig h t

S M P s, w h ic h are c o n sid e re d as a c h a ra c teristic o f th e

tra in in g p ro g ra m fo r e n g in e e rin g stu d en ts P ro b a b ilistic

m a c h in e le a rn in g m o d e ls b a se d o n th e B N N w e re

in tro d u c e d an d th e th e o re tic a l fo u n d a tio n o f th e m o d e l

a n d k e y p a ra m e te rs o f th e p ro p o se d a p p ro a c h w ere

d escrib ed , fo llo w e d b y a case stu d y u sin g a d atab ase

o f 28 9 0 sco re re su lts o f S M P s, FY P, a n d C G P A fro m a

g ro u p o f civ il e n g in eers th a t g ra d u a te d in 20 2 2 in H an o i

O n e o f th e m a in re su lts o f th e p ro p o se d m o d e l is th a t it

ca n be u tiliz e d to ev a lu a te th e in flu en ce o f an in d iv id u al

S M P o n a s tu d e n t’s final p e rfo rm a n c e in te rm s o f F Y P

a n d C G PA It w a s sh o w n fro m th e re su lts o f th e case

stu d y th a t th e su b je c ts c o m m o n ly a p p lie d in p ra c tic e also

c o n trib u te a m o re sig n ifican t influence In ad d itio n , the

B N N m o d e l fo r C G PA is c a p a b le o f p ro v id in g re la tiv e ly

clo se p re d ic tio n s w ith a m e a n re la tiv e e rro r o f 6.8%

F u rth e rm o re , th e ap p lic a tio n o f th e d a ta -d riv e n m o d el

is stra ig h tfo rw a rd as it is b u ilt b a s e d o n o p e n so u rce

lib ra rie s a n d th e u se r-frie n d ly p ro g ra m m in g lan g u ag e,

P y th o n , w ith o u t re q u irin g a n y sp e c ia liz e d so ftw are

F o r th e n e x t ste p o f th e study, o th e r m o d e ls su ch as

stra ig h t artificial n e u ra l n e tw o rk a n d d ro p -o u t n e u ral

n e tw o rk c a n also b e in c o rp o ra te d fo r c o m p a riso n p u rp o ses

F u rth e rm o re , o ne can co m p le m e n t th e d ata b a se w ith the

sco re re su lts o f all stu d en ts th a t g ra d u a te d b efo re, d u rin g ,

o r a fter th e C O V ID -1 9 p a n d e m ic to g a in an o v erall

p ic tu re to p ro p o se ap p ro p ria te so lu tio n s to im p ro v e

aca d e m ic activ itie s o f e n g in e e rin g u n iv e rsitie s, w h e re

th e p ro je c ts p la y v e ry im p o rta n t ro le fo r u n d e rg ra d u a te

students

COMPETING INTERESTS

The authors declare that there is no conflict o f interest

regarding the publication o f this article

REFERENCES

[1] T Mitchell (1997), Machine Learning , McGraw-Hill

Education, 432pp

[2] J Hu, H Niu, J Carrasco, B Lennox, F Arvin (2020),

“Voronoi-based Multi-robot autonomous exploration in unknown

environments via deep reinforcement learning”, IEEE Transactions

on Vehicular Technology, 69(12), DOI: 10.1109/TVT.2020.3034800.

[3] Y Bengio, Y LeCun, G Hinton (2015), "Deep learning",

Nature, 521(7553), pp.436-444.

[4] S.S Abu-Naser, I.S Zaqout, M Abu-Ghosh, R.R Atallah,

E Alajrami (2015), “Predicting student performance using artificial neural network: In the Faculty of Engineering and Information

Technology”, International Journal o f Hybrid Information

Technology, 8(2), pp.221-228.

[5] E.Y Obsie, S.A Adem (2018), “Prediction of student academic performance using neural network, linear regression and

support vector regression: A case study”, International Journal o f

Computer Applications, 180(40), pp.39-47.

[6] Z Iqbal, J Qadir, A.N Mian, F Kamiran (2017), “Machine learning based student grade prediction: A case study”, https://arxiv org/abs/1708.08744

[7] S.D.ABujang, A Selamat, R Ibrahim, O Krejcar, E Herrera- Viedma, H Fujita, N.A.M Ghani (2021), “Multiclass prediction

model for student grade prediction using machine learning”, IEEE

Access, 9, pp.95608-95621.

[8] S.T Jishan, R.I Rashu, N Haque, R.M Rahman (2015),

“Improving accuracy of students’ final grade prediction model using optimal equal width binning and synthetic minority over-sampling

technique”, Decis Analyst, 2(1), pp.1-25.

[9] A Polyzou, G Karypis (2016), “Grade prediction with models

specific to students and courses”, Int J Data Sci Analyst, 2(3-4),

pp.159-171

[10] I Khan, A A1 Sadiri, A.R Ahmad, N Jabeur (2019),

“Tracking student performance in introductory programming by

means of machine learning”, Proc 4th MEC Int Conf Big Data

Smart City (ICBDSC), pp.1-6.

[11] M.A Al-Barrak, M Al-Razgan (2016), “Predicting students

final GPA using decision trees: A case study”, Int J Inf Educ

Technol., 6(7), pp.528-533.

[12] E.C Abana (2019), “A decision tree approach for predicting

student grades in research project using WEKA”, Int J Adv Comput

Sci Appl., 10(7), pp.285-289.

[13] F Ahmad, N.H Ismail, A.A Aziz (2015), “The prediction

of students’ academic performance using classification data mining techniques”, 4/>/>/ Math Sci., 9, pp.6415-6426.

[14] T Anderson, R Anderson (2017), “Applications of machine learning to student grade prediction in quantitative business courses”,

Glob J Bus Pedagog., 1(3), pp 13-22.

[15] Ng Andrew (2019), Stanford University CS229: Lecture

Notes, 216pp.

[16] C.M Bishop (1997), “Bayesian neural networks”, Journal o f

the Brazilian Computer Society, 4, pp.61-68.

[17] E Bingham, et al (2019), “Pyro: Deep universal probabilistic

programming”, The Journal o f Machine Learning Research, 20(1),

pp.973-978

[18] F Pedregosa, et al (2011), “Scikit-leam: Machine learning

in Python”, The Journal o f Machine Learning Research, 12, pp.2825-

2830

D E C E M B E R 2 0 2 2 V O L U M E 6 4 N U M B E R 4 \ itliiani Jimnial nl Science,

leclnmluiiv mul Fn” incei'ing 15