International Journal of Innovative Technology and Exploring Engineering (IJITEE)
ISSN: 2278-3075, Volume-8 Issue-9, July 2019
A Model to Predict Pay Scale Fixation in Job
Market Based On Educational Excellence
Abishek kamalanathan, Archana Tamizharasan
Abstract: The students of current generation worry a lot about II. BACKGROUND
their future salary when they get employment. The salary of an
employee has always remained the major concern. In this paper Previous Researchers used statistical techniques methods to
we study and understand the major factors which are influencing predict the salaries Young-joo Lee et al. applied regression
the pay scales of the employees. This Research paper is based on method to predict the salaries considering details of student’s
predicting the pay scales of employees working in the southern academic performances and comparing the satisfaction with
part of India. The pay scale of an employee with below 4 lakh per
annum is treated as medium pay scale and pay scale with above 4
salary [2].P. khongchai et al.[4] build a salary predicting
lakh per annum is treated as good pay scale. In this paper the system using k-NN, SVM used to predict the pay scales. This
machine learning model is applied to predict the pay scale of an paper consider factors like gender,gpa,branch of study.[3]
employee with the features which best suits the model. The year of used multiple regression approach to predict the salaries of
passing, English marks, Quant marks, logical marks scores in two groups of one who are Graduates and the other groups are
AMCAT’s exam, programming skills and college CGPA were the
drop out and compared the results of expected salaries by the
major factors influencing the pay scales. Three machine learning
algorithms were applied to the dataset Naïve bayes, Decision tree adults and the actuals salaries of the adults.[9] This paper
and Bagging .Naive bayes and bagging model gave the best used statistical approach and considers factors like GMAT
accuracy of 85.2% scores, age, Gender to Predicted and analyse how the salaries
differ from the students graduating from the MBA programs.
Index Terms: Feature Selection, job market, pay scale forecast, Most of the research works in this field were related to
predicting salaries
salaries hence they followed regression method to predict the
continuous variable. This paper uses classification approach
I. INTRODUCTION
by binning the continuous values of salaries into categorical
The broad mind set of almost every student is that they values. Hence making classification of pay scales and other
believe good CGPA will help them to get good salaries. This control variables which are contributing to build a good
mind set of the students prompted me to take to this topic and model were made into categorical from continuous variable.
understand the factors which really have an impact in the Classification method can provide more deeper
salary apart the CGPA. In this paper we found that solely understanding ,insights and helps us to analyse with different
CGPA cannot influence the pay scales of an employee though machine learning model than the regression method .Machine
CGPA had an impact in the model but factor like experience, learning model like Naïve Bayes, Bagging accept categorical
marks obtained in AMCAT’s examination and programming values than continuous values. Bagging is a type ensemble
skills influenced more than the CGPA. The salary predicting which aggregates all the results and votes for the best
model will help the upcoming students to analyse about the predicted outcomes for each test data which gave the best
salary trends in job market and will actuate the students accuracy. There were 37 control variables in the dataset which
towards the right factors which will influence them to get were fed into random forest model for feature selection to
good salaries [1]. This dataset was obtained from AMEO. subset the importance factors which contribute to the model.
This dataset consist of 37 independent features out of which Surprisingly the results we got that the academic
after data pre-processing and feature selection only 6 performances in the schools did not have an impact in the
significant attributes were selected passout_year, eng_mark, model. This is was not done by other researchers as far our
quant_mark, prog_skills, cgpa_clg and logic_mark. The knowledge.
entire coding was done in R. The dataset was split into train
and test model in the ratio 80 and 20.The predictor labels are III. DATASET DESCRIPTION
medium pay and good pay, the model performed well for the
This dataset was obtained from Aspiring Minds’
medium class labels. Three models were used in this paper
Naïve Bayes, Decision Tree, Bagging.
.
Revised Manuscript Received on July 05, 2019.
Abishek kamalanathan, Computer Science, Vit University,Vellore ,
India
Archana Tamizharasan,, Computer Science, Asst. Professor,
Vit University, Vellore ,India
Published By:
Retrieval Number I7831078919/19©BEIESP
Blue Eyes Intelligence Engineering
DOI:10.35940/ijitee.I7831.078919 909 & Sciences Publication
A Model to Predict Pay Scale Fixation in Job Market Based On Educational Excellence
Attribute name Description
Salary The salary of employee INR per annum
DOJ The date of joining of employee in the company
DOL The Date of leaving of employee from the company
Designation The designation of the employee in the company
Job city The city where the employee works
Gender The Gender of the employee
DOB The date of birth of an employee
10 percentage The mark obtained by the employee in his 10th standard
10 board The board in which the employee studied in 10 standard
12 graduation The year in which the employee graduated
12 percentage The marks obtained by the employee in 12th standard
12 board The board in which the employee studied in 12 standard
CollegeID The college ID of the employee
CollegeTier The college Tier of the employee
With value 1 and 2.Tier 1 college has better standards as compared to Tier 2 colleges
Degree The degree he has completed
Specialization The department which she studied
CollegeCGPA The cgpa the employee obtained
CollegeCityTier The Tier of the college in that city
CollegeState The state in which the employee studied his college
Graduation Year The graduation of an employee
English The marks scored by the employee in
Logical the AMCAT exam
Quant
Domain
ComputerProgramming The marks scored by the employees in their respective domain stream in AMCATS
ElectronicsAndSemicon exam
ComputerScience
MechanicalEngg
ElectricalEngg
TelecomEngg
CivilEngg
Conscientiousness The score obtained by the employee in the AMCATS’S personality test.
Agreeableness
Extraversion
Neuroticism
openess_to_experience
Table i.) Table Description
0 if is not a capital city. In the box plot in Fig:1 a. shows that
Employability Outcomes (AMEO).This dataset has the salaries of employee working in capital City is greater than the
details of the employees from their 10th standard marks to non-capital City. The boxplot Fig: 1b. shows the comparison
college CGPA. The dataset before pre-processing consisted of salaries between male and female, there was no difference
of 37 features. in mean salaries between male and female.Pay scale attribute
The description of each feature is given in the table i. was introduced and the salaries below 4 lakhs were labelled as
medium pay and salaries above 4 lakhs were labelled as Good
IV. DATA PREPROCESSING pay The continuous attributes English,
The attribute Jobcity had many ambiguous spelling for the
cities like Chennai, Madras and Banglore, Bengaluru which
was corrected and had one correct spelling for all the City.
The College Cgpa attribute had CGPAs calculated for 100
and some were calculated for 10.Since most of the colleges
had CGPAs out of 100 all CGPAs were scaled to 100.The
continuous factor like Graduation year, English, Logical,
Quant, College CGPA, Computer Programming were binned
into categorical values. A new attribute IsCapital attribute was
introduced this attribute has values 1 if it is capital city
(Chennai, Bengaluru, Hyderebad, Thiruvananthapuram) and
Published By:
Retrieval Number I7831078919/19©BEIESP
Blue Eyes Intelligence Engineering
DOI:10.35940/ijitee.I7831.078919 910 & Sciences Publication
International Journal of Innovative Technology and Exploring Engineering (IJITEE)
ISSN: 2278-3075, Volume-8 Issue-9, July 2019
medium class labels and only 219 records of good class
labels, the number of training examples of good class labels
were less compared to the medium class labels which make
the model difficult to predict for good pay labels. The Fig 2
and 3 below shows the variable importance and measure of
(MDA) and MeanDecreaseGini.
VI. PROPOSED MODEL
The dataset obtained after pre-processed is used for building
the machine learning model. The target attribute has two
values medium pay and good pay which is the main objective
Fig 1 a.) of this paper to classify the employees pay range.
Three machine learning algorithm was used in making the
model. The control variables are selected from the Feature
Selection, so the features which contribute significantly in
building the model are passout_year, eng_mark, quant_mark,
prog_skills,cgpa_clg,logic_mark. The Fig4 depicts the model
flow.
Fig 1 b.)
Logical, Quant, Domain were converted into new attributes
English mark, Logical mark, Quant mark, Domain mark with A. Naïve Bayes
categorical values having labels >=500 or <500.The Naïve Bayes algorithm is one of the best algorithms for classi
Graduation Year attribute was binned into passout_year with fication problems with categorical values. It is based on the
values 2007-2011 and 2012-2016. The attribute Computer Bayes probability theorem works with an assumption that the
Programming has continuous variable of marks scored in probability of a feature does not affect the other feature that
AMCAT’s programming examination which is binned in a is the probability of the feature given the predictor is indepe
new attribute Prog_skills with values >=425 and <425. The ndent of other features.
attribute College_CGPA had continuous variables which was
categorised as >=70 and <70 in a new attribute cgpa_clg.
V. FEATURE SELECTION
The feature Selection was done by using Random Forest
model. The package RandomForest in R was used where the
function importance () gives the measure of Mean Decrease
Accuracy and Mean Decrease Gini of each attributes. The
attribute with larger value of Mean Decrease Accuracy
(MDA) is more important feature for the model. The value of
MDA tells the decrease Accuracy (MDA) is more important
feature for the model. The value of MDA tells the decrease in
the accuracy if the attribute is removed from the feature list.
The Gini index is the measure of the purity or the
homogeneity of the attribute contributing towards the split of
the node. A low Gini has higher Decrease Gini thus that
attribute contributes to the split of the node. The attributes
IsCapital, Gender, 12th board, domain range do not have
significance in the model. The attributes which has values less
than zero in mean decrease accuracy is not considered in
building the machine learning model. The attribute
passout_year represents the experience of an employee was
the most significant attribute which becomes the root node in
the decision tree. The pre-processed data had 1219 records of
Published By:
Retrieval Number I7831078919/19©BEIESP
Blue Eyes Intelligence Engineering
DOI:10.35940/ijitee.I7831.078919 911 & Sciences Publication
A Model to Predict Pay Scale Fixation in Job Market Based On Educational Excellence
Fig 2.) The variable importance of each attribute for feature selection
Fig 3.) The value of MeanDecreaseAccuracy and MeanDecreaseGini
B. Decision Tree
The prior probability of the medium pay class label is 0.78 a Decision Tree is one of the best algorithm for classification.
nd prior probability of good pay class label is 0.21.The condi Decision trees can handle both continuous variables and
tional probability of each attribute is given in Table ii.). The categorical variables in-case of continuous variables it
model had an accuracy of 85.2%. becomes regression tree and in the other case it becomes
classification tree. Classification works well with datasets
Class Attribute with limited number of records. Classification tree works on
the entropy approach. The attribute passout_year has the
Feature Attribute Medium highest information gain so it becomes the root node. The
Good Pay
Pay marks obtained in the quant has the second greatest
information gain value, it formed the next split after the root
passout_ye 2007-2011 0.1523438 0.4676259 node. The cgpa_clg and programming_skills formed the
ar 2012-2016 0.8476562 0.5323741 subsequent splits respectively. The information gain is the
difference between the entropy of the class attribute and
<500 0.5585938 0.4388489 entropy of the feature.
eng_mark
>=500 0.4414062 0.5611511 The entropy of attribute is calculated as
quant_mar <500 0.5517578 0.3093525
k >=500 0.4482422 0.6906475
<425 0.5566406 0.3920863 From the decision tree it can be inferred that The probability
prog_skills of a employee being in good pay is high if passes out in year
>=425 0.4433594 0.6079137
<70 0.3789062 0.2913669
cgpa_clg
>=70 0.6210938 0.7086331
logic_mar <500 0.5439453 0.4388489
k >=500 0.4560547 0.5611511
Table ii.) Conditional probability of each feature
Published By:
Retrieval Number I7831078919/19©BEIESP
Blue Eyes Intelligence Engineering
DOI:10.35940/ijitee.I7831.078919 912 & Sciences Publication
International Journal of Innovative Technology and Exploring Engineering (IJITEE)
ISSN: 2278-3075, Volume-8 Issue-9, July 2019
Fig:5.The decision tree showing the node splits the right side split denotes YES and left side split denotes NO
2007-2011 and his quant mark is greater than 500 and cgpa is the employees programming skills and his college CGPA.
greater than 70. Since bagging has better accuracy, the best model is chosen as
If an employee has graduated from 2011-2016 then it comes bagging. In the test data there were 255 instances of medium
under the category of medium pay, also if the employee does class labels out which the bagging model predicted 246
not have good marks in quant and programming skills is more instances correctly and Decision tree predicted 244 instances
likely to be in medium pay scale. Even if the cgpa of an correctly and the test data had only 52 good class labels out
employee is less than 70 and his programming skills is good which the bagging model could able to predict the 22
he is likely to be having a good pay scale. The accuracy of the instances correctly and Naïve Bayes predicted only 18 and
model was 84.7%.The visualization of decision Tree is shown decision Tree predicted 19. In both class labels bagging
in Fig 5. model had an edge over other two models. The comparison is
shown in Table iii
C. Bagging
Bagging is a type of ensemble learner. Ensemble classifier
combines various algorithms and provides the best solution. It
is a combination of bootstrap and aggregation were bootstrap
creates a random sample data from the train data and
aggregate all the outcomes from the model. Each model votes
for the test instances, the vote here is predicting the class
labels. The maximum votes get elected and becomes the final Table iii.) comparison of algorithms used.
outcome of the model. The accuracy of model was 85.9%..
REFERENCES
VII. CONCLUSION
1. P. Khongchai, P. Songmuang, "Random Forest for Salary Prediction
The model performed well for the Medium pay Labels as it System to Improve Students? Motivation", 016 12th International
Conference on Signal-Image Technology & Internet-Based Systems
had more training labels. The model showed no difference in (SITIS), pp. 637-642, 2016.
pay scales with respect to Gender, also the marks obtained in 2. Y. Lee, M. Sabharwal, "Education—Job Match Salary and Job
schools were insignificant in building the model. There was a Satisfaction Across the Public Non-Profit and For-Profit Sectors:
significant decrease in the number of employee who scored Survey of recent college graduates", Public Management Review, vol.
18, no. 1, pp. 40-64, 2014.
good marks in class 10 when compared to class 12. The mean 3. J. Jerrim, "Do college students make better predictions of their future
salary of employee who holds B.Tech degree was higher than income than young adults in the labor force?", Education Economics,
the M.tech. The mark obtained in the respective domain exam vol. 23, no. 2, pp. 162-179, 2013.
4. P. khongchai, P. Songmuang, "Improving Students'Motivation to
was insignificant in predicting the mode. The employees Study using Salary Prediction System", 13th International Joint
working in the capitals cities had large number of outliers than Conference on Computer Science and Software Engineering
the employees working in non-capital areas. The personality (Preprint), 2016.
5. T. G. Dietterich, "An Experimental Comparison of Three Methods for
test(extraversion,Agreeableness,neuroticism,openess_to_exp Constructing Ensombles of Decision Trees: Bagging Boosting and
erience and agreeableness) marks had no impact in the model. Randomization", Machine Learning, vol. 40, no. 2, pp. 139-157, 2001.
The college tier had no significance in deciding the pay 6. Betts, J. “What Do Students Know About Wages? Evidence from a
Survey of Undergraduates.” Journal of Human Resources 31 (1):
scales. The experience of the employee was the important
27–56 doi: 10.2307/146042, 1996
factor in deciding the pay scales. The marks obtained in 7. Jerrim, J “The Wage Expectations of UK Students: Are They
logical, quant, computer programming and college CGPA had Realistic?.” FiscalStudies 32
an impact in predicting the pay scales. It can be concluded that (4): 483–509 doi:
the employee with more experience likely to get a good pay
scale and other major factors influencing the pay scales are
Published By:
Retrieval Number I7831078919/19©BEIESP
Blue Eyes Intelligence Engineering
DOI:10.35940/ijitee.I7831.078919 913 & Sciences Publication
A Model to Predict Pay Scale Fixation in Job Market Based On Educational Excellence
10.1111/j.1475-5890.2011.00148,2011
8. Karla R. H. and W. A. Hamlen, “Faculty salary as a predictor of student
outgoing salaries from MBA programs", Journal of Education for
Business, 91:1, p 38-44, 2015
9. B. Xu, J. Li, Q. Wang, X. Chen, "A Tree Selectllion Model for
Improved Random Forest", Bulletin of Advanced technology research,
vol. 6, no. 2, 2012
10. O. Villacampa, "Feature Selection and Classification Methods for
Decision Making: A Comparative Analysis", College of Engineering
and Computing Nova Southeastern University, 2015.
11. Guyon I, Elisseeff A: An introduction to variable and feature
selection. JMachLearnRes,3:1157–82.10.1162/15324430332275361
6, 2003
12. Archer, K.J., Kimes, R.V.: Empirical characterization of random
forest variable importance measures.Comput.Stat. Data Anal. 52,
2249–2260, 2008
13. Biau, G., Devroye, L., Lugosi, G.: Consistency of random forests and
other averaging classifiers. J. Mach. Learn. Res., 2015–2033,2008
14. Dietterich, T.G.: An experimental comparison of three methods for
constructing ensembles of decision trees: Bagging, boosting, and
randomization. Mach. Learn. 40, 139–157,2000.
AUTHORS PROFILE
Abishek kamalanathan, He is pursuing his 4th
year computer science in Vellore Institute of
Technology. He is an Machine learning and data
science enthusiast. He has done his academic
internship on Data Science in National University
of Singapore. His aspiration is do Masters in data
science and purse research works data science and
machine learning.
Archana Tamizharasan is an assistant professor in
School of computer science and engineering at
Vellore Institute of Technology. She is pursuing her
Ph.D. She has publications in interdisciplinary
research on Market analysis in e-business and
psychological pattern mining in virtual learning
environment. Her current research focus is
exploratory data analysis and applications of deep
and machine learning.
Published By:
Retrieval Number I7831078919/19©BEIESP
Blue Eyes Intelligence Engineering
DOI:10.35940/ijitee.I7831.078919 914 & Sciences Publication
