See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/341778658

Predicting Students’ Employability using Support Vector Machine: A SMOTE-

Optimized Machine Learning System

Article · May 2020

DOI: 10.30534/ijeter/2020/102852020

CITATIONS READS

27 2,916

3 authors:

Cherry Casuat Enrique Festijo

Technological Institute of the Philippines Technological Institute of the Philippines
21 PUBLICATIONS 219 CITATIONS 46 PUBLICATIONS 333 CITATIONS

SEE PROFILE SEE PROFILE

Alvin Sarraga Alon

National Research Council of the Philippines
53 PUBLICATIONS 639 CITATIONS

SEE PROFILE

All content following this page was uploaded by Alvin Sarraga Alon on 31 May 2020.

The user has requested enhancement of the downloaded file.

 ISSN 2347 - 3983
Volume 8.
Cherry D. Casuat et al., International Journal of Emerging No. 5,
Trends May 2020 Research, 8(5), May 2020, 2101 - 2106
in Engineering
International Journal of Emerging Trends in Engineering Research
Available Online at http://www.warse.org/IJETER/static/pdf/file/ijeter102852020.pdf
https://doi.org/10.30534/ijeter/2020/102852020

Predicting Students’ Employability using Support Vector

Machine: A SMOTE-Optimized Machine Learning System
Cherry D. Casuat1, Enrique D. Festijo2, Alvin Sarraga Alon3
1
Technological Institute of the Philippines, Philippines, ccasuat.cpe@tip.edu.ph
2
Technological Institute of the Philippines, Philippines, enrique.festijo@tip.edu.ph
3
Technological Institute of the Philippines, Philippines, aalon.cpe@tip.edu.ph

 1. INTRODUCTION
ABSTRACT
One of the main challenges of schools and universities today
The graduates in every institution reflect the skills developed is providing programs that are aligned with the policy of
and competencies acquired by the tudents through the commission on higher education (CHED) and delivering
education offered by the institution that is suitable in the consistent outcomes that can be accepted not only in the
companies. Employability of graduates becomes one of the Philippines but also in other countries. The higher educational
performance indicators for higher educational institutions institutions shift their paradigm from just simply educating
(HEIs). Therefore, it is important to accentuate the the students to developing life-long learning skills. The
employability of graduates. This is the reason why this CHED in the Philippines initiated reforms for education
research is being carried out. This study involved sectors through the conversion of the traditional way of
twenty-seven thousand (27,000) information consist of three teaching known as teacher-centered to an outcomes-based
thousand (3000) observations and twelve (12) features of education (OBE) curricula that are commonly known as
student’s mock job interview evaluation results (MJI), student-centered, as per CHED-CMO No. 46 s. 2012 [1]. The
on-the-job training (OJT) student’s performance rating and higher education institutions’ programs offering and syllabus
general point average (GPA) of students enrolled in the were based on the principles of OBE by which accentuate the
on-the-job training course of SY 2015 to SY 2018. To address type of delivery of its services to the students. To strengthen
the issue in imbalance datasets where the minority class, the and assist the country by producing graduates with critical
researchers used synthetic minority over-sampling technique thinking, behavioral and life-long learning skills and
(SMOTE) were applied in this study to address the issue in competencies aligned with institutional learning outcomes,
imbalanced datasets Six learning algorithms with SMOTE industry desired values, and international standards [2].
were used such as Decision Trees (DT), Random Forest (RF),
and Support vector machine (SVM), K- Nearest Neighbor Graduate readiness study has also been conducted by other
(KNN), Logistic Regression (LR) to understand how students, scholars, where the aim of their paper is,(1) to investigate
get employed. The six algorithms were evaluated through the student’s experience when it comes to essential skills acquired
performance matrix as accuracy measures, precision and in the university ; (2), student perception when it comes to
recall measures, f1-score, and support measures. During the potential job role when they graduate [3]. According to a
experiments, Support Vector Machine (SVM) obtained recent survey by the Malaysian Ministry of Education
91.22% inaccuracy measures which were significantly better assessing “(youth)” unemployment and graduates, In
than all of the learning algorithms, DT 85%, RF 84%. The particular, just 53 percent of the 273,373 graduates find jobs
learning curve produced during the experiment displays the within six months of graduation in 2015, 24 percent of
training error results which were above the one for validation graduates have no job after graduation and 18 percent were
error while the validation curve displays the testing output engaged in continuing education. That is why, as mentioned,
where gamma was best at 10 to 100 in gamma 5. This only 53% of students were employed because of “the
concludes that the model produced with SVM was not under discrepancy between the education provided at the
fitted and over-fit. This study is very promising which leads universities and the skills needed by the industry. According
the researchers to be motivated to enhance the process and to to the researchers in Malaysia and China, most of the
validate the produced predictive model for further study. university curriculum where they conducted their studies
reflect the current skill requirements of the industry [4]. In a
Key words : Employability prediction system, Decision trees, Alsore researchers have studied graduate employability.
K-nearest neighbour, Logistic regression, Naïve Bayes, Research utilizing data mining and modeling methods has
Random Forest, SMOTE, Support Vector Machine been carried out that highlights particular computing and data
management challenges. [5]. Using the data analytics, the
results were evaluated by (1) monitoring the job status of
graduates by giving them prompts and invitations; (2)
encouraging them to maintain track of the position they

2101
Cherry D. Casuat et al., International Journal of Emerging Trends in Engineering Research, 8(5), May 2020, 2101 - 2106

wanted; and (3) determining the jobs approaches function 2.1 Contextual Diagram
well, especially in the sector-specific region. [6].
Unfortunately, the complexity of the workplace and the
advent of modern technologies have shifted, and the varied
demands. of consumers demand the definition of “globally
competitive” for employable graduates. It also challenges the
capacity of universities to satisfy the need for graduates who
are suitable for the job in the industry [7]. These concepts of
career development and employability lead college education
to evaluate their program offerings and to test their
effectiveness and congruence with the needs of the sectors to
deliver qualified students who can quickly be consumed by
industries.

The assessment of being suitable for the job often follows the Figure 1: Contextual Diagram
theory of human capital, in which individuals’ personal and
technical growth are called assets in human capital that acts as This figure 1 shows how the system intends to work in
factors for their degree of employment and personal earnings. predicting the employability of undergraduate students. First,
Graduates will then make substantial improvements in their it is important to know the objective is to predict the student’s
intellectual resources to allow their potential employers more employability. Next is compiling all the student’s data, in this
marketable [8]. However, It is increasingly important for stage the datasets collected will be cleaned and normalized
individuals to retrain or develop new knowledge and abilities and then merged different datasets. The datasets will be
to address the requirements of the rapidly growing workforce trained using the six algorithms, the best model that will be
and the multifaceted entrepreneurial environment. They will created will be used in the system that will be developed.
develop their professional abilities, values, and work 2.2 Datasets Collection
experience and adapt to the changing labour market
Table 1: Students’ Employability Datasets
requirements.

Most published researches and studies used data mining

techniques to predict employability. Some of the techniques
were Tree of Decision, Naïve Bayes, and Vector Machine
Support [9]. Often used in data mining techniques are the
Logistic Regression, K-Nearest Neighbor, Random
Woodland, SVM (Linearsvc), Quadratic Discriminant
Analysis (QDA), and Multi-class Ada Boosted [10]. The
application of Machine learning when it comes to forecasting
employability is in the infancy period. They compared
numerous algorithms in the analysis carried out by Ohio
University where the datasets used were from business
education. The aforementioned research will not find the
datasets regarding the mismatch [11].
The datasets were collected from different agencies in the
This paper seeks to establish a machine learning method to university which consists of Mock job Interview Results
forecast the employability of the applicant and to examine the consist of three thousand (3000) observations and twelve (12)
signs of their skill set. This paper is in the production stage of features, Student Performance Rating of the OJT students
a model focused on machine learning to forecast the collected by the On-The-Job Training (OJT) Faculty
employability of students. The researchers were inspired to In-charge and General Point Average from the Registrar’s
perform the study in the light of emerging areas such as Office. The datasets collected need to be normalized and
operational intelligence or instructional analytics to cleaned. The datasets that were collected were compliant with
strengthen and encourage certain ability sets found that will the Data Privacy Act of the Philippines.
lead to the enhanced jobs of engineering students.
2.3 Preprocessing of Datasets
2. METHODOLOGY
The preprocessing stage consists of cleaning the first of the
datasets. The researchers used data normalization where each
The methodology of this proposed method was divided into
attribute or column was filled with the median values when
three-phase such as Data collection, Preprocessing, Training
there is a missing value on attributes or columns. For row or
of datasets which handling imbalance datasets were
number of observations were filled with the mean of that
highlighted in the study.
number of observations or row when there is a missing value
for row [10]. Then merging of the cleaned datasets to create a

2102
Cherry D. Casuat et al., International Journal of Emerging Trends in Engineering Research, 8(5), May 2020, 2101 - 2106

consolidated dataset that comes from a mock-job interview, it as CSV file directly on the documents folder of your
OJT student performance rating, a general percentage grade computer. PYQT5 and QT designer will also be used for the
GUI design
2.4 Training the Datasets
The proponents trained the datasets using the 70-30 splitting 2.5.2 Functional View
of datasets. The learning algorithms such as SVM, Decision
Trees, Random Forest, Logistic Regression, Naïve Bayes, and
KNN. Based on the training conducted the SVM got the
highest accuracy of 92.22% of the entire used algorithm,
which means that the Support Vector Machine was the
best-created model. The Support Vector Machine (SVM)
analyzes the data for classification analysis.

2.2.1Handling Imbalanced Datasets

The synthetic minority over-sampling technique (SMOTE)
which applies the k-nearest neighbor algorithm that chose,
combines and generates the synthetic samples in the nearest
space. The algorithm takes the vectors of the attributes and its
closest neighbors, measures the difference between such
paths. It is multiplied and added back to the feature by a Figure 3: The Proposed System Functional View
random number (0, 1). SMOTE algorithm is a pioneer and
SMOTE is a basis for several other algorithms [12]. In this Figure 3 shows the proposed system input, process, and
study, SMOTE was used to addressed the issues of output views. Mock-job interview .csv and OJT assessment
imbalanced datasets in employability datasets where the tool Ratings will be accepted for input. Then, once cleaned
majority class is employable. and merged, the SVM model will be applied to predict the
employability of the students. There are different studies
2.5SOFTWARE DESIGN applied prediction such as in career management [13], also in
shortlisting of job [14]and some are for modelling purposes
[15] – [17].

2.5.3 System Flowchart

Figure 2: System Architecture

Figure 2 shows that the first step is to input data. The datasets
will be preprocessed (cleaned and merged, splitting of
datasets into 70,30). Then, training of datasets will take place
to create a model. One model was created, that model will be Figure 4: Students’ employability prediction system flow chart
used in the system. The user can log-in to the system GUI by
providing a password and log-in successfully. Only those who The figure above shows how the data flows and the decision
have a user account can upload the datasets and predict if the was made to control the events. The process will apply the
list of students is employable or less employable. algorithm that was chosen. After the pre-processing, the
analyzed dataset will split into three categories which are
2.5.1 Constraints testing, training, and validation. After the training and using
The system will predict student’s employability if it was the model, the system will now predict and show the accuracy
merged in the On-the-Job Training CSV file and Mock Job result based on the datasets.
CSV file at a time. The system will use acquired machine
learning techniques using Python programming language to
process the data that will show the result in the GUI and save

2103
Cherry D. Casuat et al., International Journal of Emerging Trends in Engineering Research, 8(5), May 2020, 2101 - 2106

3 TESTING AND VALIDATION

3.4 System Graphical User Interface
3.1 Performance Measures The best model created which is the SVM was applied to the
Table 1: Students’ Employability Datasets employment prediction system.

3.4.1 User’s account registration

Among the learning algorithms in table 2, SVM obtained the

highest accuracy which is 91.22%, 91.10% for recall, and
both 91% for f1-score and precision.

3.2 Learning curve

Figure 7: Account registration

The figure above shows the account registration where the

user will create an account first.

Figure 5: Support Vector Machine learning curves in gamma 5

Figure 5 shows the learning curve of SVM where the values in

maximum training score mean is 0.960843, the maximum
cross-validation score means is 0.850478, maximum training
score is 0.9819277, maximum cross-validation score is
0.72966. The learning curve for the training error results was Figure 8: Log-in Interface
above the one for the validation error. The accuracy measure
described how good the model is and the MSE on the other Figure 8 shows the log-in interface of the system. The user
side described how bad the model is. The irreducible error log-in was his/her credentials to be able to use the system.
gives an upper bound.
3.4.2 Uploading and merging of datasets
3.3 Validation Curve

Figure 6: SVM Validation Curve in gamma5

Figure 6 shows the validation curve with SVM in gamma 5

where the maximum training R-squared score was 0.918 and Figure 9: Uploading the Mock job and OJT datasets
the maximum cross-validation score was 0.857. It shows that
the validation curve with SVM in gamma 5, the gamma is best
at 10 to 100.
2104
Cherry D. Casuat et al., International Journal of Emerging Trends in Engineering Research, 8(5), May 2020, 2101 - 2106

3.4.3 Student’s Employability Result researchers concluded that the learning curve and validation
curve that it showedwas not overfitted or underfit.
ACKNOWLEDGEMENT

The proponents would like to thank the Career Center of

TIP-Manila especially to the SDP Officer and Career Adviser
for their unwavering support to the proponents and the MR.
SUAVE Laboratory of Technological Institute of the
Philippines for all computing facilities that researches have
been used to make this study possible.

REFERENCES
Figure 10: Students’ employability prediction resultswhere the
SVM model was applied 1. CMO 46 s. 2012 - CHED, CHED, 2020. [Online].
Available: https://ched.gov.ph/cmo-46-s-2012/.
Figure 10 shows the application of the SVM model in 2. Implementing Rules and Regulations of the Enhanced
predicting student’s employability. The system predicts if the Basic Education Act of 2013 | GOVPH, Official
student is employable or less employable. Then the system Gazette of the Republic of the Philippines, 2020.
recommends what areas the student needs to improve to be [Online]. Available:
more employable at the time of graduation. https://www.officialgazette.gov.ph/2013/09/04/irr-repub
lic-act-no-10533/.
3. W. Teng, C. Ma, S. Pahlevansharif and J. Turner,
Graduate readiness for the employment market of the
4th industrial revolution, Education + Training, vol.
61, no. 5, pp. 590-604, 2019. doi:
10.1108/et-07-2018-0154
4. M. Alias, G. Sidhu and C. Fook, Unemployed
Graduates’ Perceptions on their General
Communication Skills at Job Interviews, Procedia -
Social and Behavioral Sciences, vol. 90, pp. 324-333,
2013. doi: 10.1016/j.sbspro.2013.07.098
5. B. Tapado, G. Acedo and T. Palaoag, Evaluating
information technology graduates employability
Figure 11: Excel file generated Prediction Results
using decision tree algorithm, Proceedings of the 9th
International Conference on E-Education, E-Business,
Figures 7-11 shows the developed student's employability E-Management and E-Learning - IC4E '18, 2018. doi:
prediction system. The users need to just log-in and upload 10.1145/3183586.3183603
the datasets needed to predict the employability of the 6. R. Bridgstock and D. Jackson, Strategic institutional
students. approaches to graduate employability: navigating
meanings, measurements and what really
4.CONCLUSION matters, Journal of Higher Education Policy and
Management, vol. 41, no. 5, pp. 468-484, 2019. doi:
10.1080/1360080x.2019.1646378
The Higher Education Institutions (HEIs) becoming more 7. R. Bringula, A. Balcoba and R. Basa, Employable Skills
accountable for student’s career outcomes and as jobs in the of Information Technology Graduates in the
labor market increases its competition, the institution needs to Philippines, Proceedings of the 21st Western Canadian
identify students’ employability. This study develops a Conference on Computing Education - WCCCE '16,
student employability prediction system using an SVM 2016. doi: 10.1145/2910925.2910928
machine learning approach of predicting students’ 8. L. Almendarez, Human Capital Theory: Implications
employability where the issues in imbalanced datasets have for Educational Development in Belize and the
been addressed using SMOTE. The best algorithm that has the Caribbean, Caribbean Quarterly, vol. 59, no. 3-4, pp.
highest accuracy and has the highest performance evaluation 21-33, 2013. doi: 10.1080/00086495.2013.11672495
compare to the other five learning algorithms that have been 9. W. Fok et al., Prediction model for students' future
trained to create the best model. Therefore, researchers development by deep learning and tensorflow
concluded that Support Vector Machine (SVM) produces a artificial intelligence engine, 2018 4th International
predictive model that obtained 91.22% for the accuracy and Conference on Information Management (ICIM), 2018.
for recall measures which are .911 or 91.10% and 91% for doi: 10.1109/infoman.2018.8392818
precision respectively. The researchers realized that gamma is 10. Y. Bharambe, N. Mored, M. Mulchandani, R.
best at 10 to 100 in gamma 5 as shown in figure 6. The Shankarmani and S. Shinde, Assessing employability of

2105
Cherry D. Casuat et al., International Journal of Emerging Trends in Engineering Research, 8(5), May 2020, 2101 - 2106

students using data mining techniques, 2017

International Conference on Advances in Computing,
Communications and Informatics (ICACCI), 2017. doi:
10.1109/icacci.2017.8126157
11. A. Farahat, A. Elgohary, A. Ghodsi and M. Kamel,
Greedy column subset selection for large-scale data
sets, Knowledge and Information Systems, vol. 45, no. 1,
pp. 1-34, 2014. doi: 10.1007/s10115-014-0801-8
12. N. Chawla, K. Bowyer, L. Hall and W. Kegelmeyer,
SMOTE: Synthetic Minority Over-sampling
Technique, Journal of Artificial Intelligence Research,
vol. 16, pp. 321-357, 2002. doi: 10.1613/jair.953
13. S. J, Career Prediction through Cognitive Models
using Sudoku Game – The Assessment of
Applicability, International Journal of Emerging Trends
in Engineering Research, vol. 7, no. 11, pp. 473-480,
2019. doi: 10.30534/ijeter/2019/127112019
14. R. Gustilo, An Analytic Hierarchy Process Approach
in the Shortlisting of Job Candidates in
Recruitment, International Journal of Emerging Trends
in Engineering Research, pp. 333-339, 2019.
doi: 10.30534/ijeter/2019/17792019.
15. A. Alon, A Machine Vision Detection of Unauthorized
On-Street Roadside Parking in Restricted Zone: An
Experimental Simulated Barangay-
Environment, International Journal of Emerging
Trends in Engineering Research, vol. 8, no. 4, pp.
1056-1061, 2020.
doi: 10.30534/ijeter/2020/17842020
16. A. Alon, Machine Vision Recognition System for
Iceberg Lettuce Health Condition on Raspberry Pi
4b: A Mobile Net SSD v2 Inference
Approach, International Journal of Emerging Trends in
Engineering Research, vol. 8, no. 4, pp. 1073-1078,
2020.doi: 10.30534/ijeter/2020/20842020

2106

View publication stats