Global Journal of Computer Science and Technology: D

Neural & Artificial Intelligence

Volume 19 Issue 4 Version 1.0 Year 2019
Type: Double Blind Peer Reviewed International Research Journal
Publisher: Global Journals
Online ISSN: 0975-4172 & Print ISSN: 0975-4350

Employee Culling based on Online Work Assessment through

Machine Learning Algorithm
By Khaled Redwan, Yeasin Ahammed, Masum Akram Hridoy, Fernaz Narin Nur
& A. H. M. Saiful Islam
Notre Dame University
Abstract- Job analysis, planning employee needs, recruiting the appropriate people, wages and salary
management are the important theme of human resource management. Human resource management
also includes evaluating performance, resolving problems, and create communication with all employees
at all levels. On the other hand, Machine learning is a data analytics technique that teaches computers to
do what comes naturally to humans. So through these two sectors such as computation and business
administration, in this paper on employee culling based on work assessment by which machine learning
algorithm such as KNN, SVM, The Decision tree can give the best result (perfect employee). We also
focus on the accuracy that algorithm is performing. We marked an employee through their experience,
language skills, skills, graduation, etc. we create e model by which we can get input through the
companies and give them a perfect result through their requirement assessment and machine learning
algorithm.
Keywords: machine learning, employee culling, human resource management, KNN. SVM.
GJCST-D Classification: F.2.1

EmployeeCullingbasedonOnlineWorkAssessmentthroughMachineLearningAlgorithm

Strictly as per the compliance and regulations of:

© 2019. Khaled Redwan, Yeasin Ahammed, Masum Akram Hridoy, Fernaz Narin Nur & A. H. M. Saiful Islam. This is a
research/review paper, distributed under the terms of the Creative Commons Attribution-Noncommercial 3.0 Unported License
http://creativecommons.org/licenses/by-nc/3.0/), permitting all non-commercial use, distribution, and reproduction in any
medium, provided the original work is properly cited.
 Employee Culling based on Online Work
Assessment through Machine Learning
Algorithm
Khaled Redwan α, Yeasin Ahammed σ, Masum Akram Hridoy ρ, Fernaz Narin Nur Ѡ
& A. H. M. Saiful Islam ¥

2 019
Abstract- Job analysis, planning employee needs, recruiting learning by which we can provide them the best

Year
the appropriate people, wages and salary management are employees from so many employees dataset, and thus
the important theme of human resource management. Human the paper determines Employee culling based on online
resource management also includes evaluating performance, 23
work assessment through the machine learning
resolving problems, and create communication with all
employees at all levels. On the other hand, Machine learning is
algorithm. Through our Process, we can find the best

Global Journal of Computer Science and Technology ( D ) Volume XIX Issue IV Version I
a data analytics technique that teaches computers to do what employee through the skill requirement of a client and
comes naturally to humans. So through these two sectors company. The process will focus on the experience
such as computation and business administration, in this education, and the skills of an employee. This process
paper on employee culling based on work assessment by can be a future human resource management
which machine learning algorithm such as KNN, SVM, The application.
Decision tree can give the best result (perfect employee). We
also focus on the accuracy that algorithm is performing. We II. RELATED WORK
marked an employee through their experience, language skills,
skills, graduation, etc. we create e model by which we can get In this paper, we create a model but, it has
input through the companies and give them a perfect result previous literature, which is known as the “Holy Grails”
through their requirement assessment and machine learning project to AMAZON. In which they put the information of
algorithm. the employee, and the algorithm will provide the best
Keywords: machine learning, employee culling, human and talented one. But this company had some
resource management, KNN. SVM. limitations in 2015. The company have realized their
new system, which couldn’t rating their candidates in a
I. INTRODUCTION gender neutral way in the software developer jobs. The

A
ccording to Human Resource Management, the reason is the computer model have analyzed patterns
process is culling employee, give them perfect in resumes which is submitted over a 10 year period.
training and development, appreciate their And in that tech industry most of the employees are
performance, maintain relationships, ensuring employee male. At “Unilever” they select their employees through
satisfaction, health and welfare management and some process. They have to complete an application
maintain labor laws, etc. And now a day’s machine form. It is an easy process. They can apply through
learning is gaining the perfect analysis and prediction “Linked In” profile. On the application we can type many
through data and they have the automated tools which functions through varies and region. After that, they have
can analyze the profile of business user. Through the to play 12 online analytical games. After completing the
machine learning algorithm, we provide the best games they will receive personalized feedback. They
employee to a company. In the busiest era, it is time- have solve real- world problems through “Unilever”
wasting matter to find the best employee from so many scenarios. A digital interview will occur. Where a
resumes and it is a challenging matter too. The HR candidate needs a suitable internet connection.
department of any company has to put his valuable time But “Unilever” has some limitations too, they got
to find an employee by a manual process if there is shortlisted employees but didn’t get the perfect
employee turnover in that company. So based on employee. [1][9].
companies and online work requirement assessment, Now the difference between manually employee
we have developed a model and through machine selection process and employee selection through
machine earning process is bellowed [1] [9].

Author α σ ρ Ѡ ¥ : Notre Dame University Bangladesh

Dhaka, Bangladesh. e-mails: [email protected],
[email protected], [email protected],
[email protected], [email protected]

© 2019 Global Journals

 Employee Culling based on Online Work Assessment through Machine Learning Algorithm

Employee selection through machine

Step No. Manually employee selection process
learning algorithm
Identify your hiring needs Load applicants list
1.

2. Analyze the position Get input of client’s requirement

Mark an employee through his experience, skills,
3. Evaluation, description and person specs
cgpa etc
4. Develop a recruitment strategy Mark an employee through his analytical skill
5. Go through the applications Put these marks in the algorithm
Conduct interviews and tests and conduct a final
2 019

6. Get the employee according to the requirement

interview
Year

III. Methodology
24 Process 1: To find the best employee, we have to reduce
the applicant’s list through some basic requirements of
Global Journal of Computer Science and Technology ( D ) Volume XIX Issue IV Version I

the client.
Process 2: Remove those applicants who are not in the
experience zone.
Through these two processes, we can reduce
our dataset. Then the recruiter will provide some
questions or games, and the workers have to answer it
or should use their analytical skills. This exam will provide
marks (50%)
The second stage is, the employee have to
submit a video interview, where the assessor is not a
human but a machine. The machine examines the
candidates who has to answer questions for around 30 Figure 1: Graph 1
minutes. The natural language processing and body Now the client will input the value of marks and
language analysis will determines who is fit for the job. potentiality of the employee that he needs for his
This interview will provide the value of potentiality (50%) company.
[4][7]. Let the client need some of the best employees
Total number= Mark (50%) + potentiality (50%) = 100% whose marks are = 40 and potentiality is= 40. And he
wants to show four best employees out of 10 employees.
Now a data set is given bellow
Here k = 4 and (x, y) = (40, 40)
Employee
Number Marks Potentiality Now, we have to put the point (40, 40) on the GRAPH 2
name
1 Jack 15 20
2 David 20 20
3 Jason 25 22
4 Francis 30 25
5 Chester 32 23
6 Marlin 35 32
7 Shuvashish 40 38
8 Arthur 42 39
9 Mary 43 40
10 Samuel 46 50

Dataset 1 Figure 2: Graph 2

Put the value GRAPH 1 where Here X-axis contains the
value of Marks and Y-axis contains Potentiality
© 2019 Global Journals
 Employee Culling based on Online Work Assessment through Machine Learning Algorithm

Now the K nearest neighbor algorithm will:

■ Load the data
■ Initialize the value = k
■ For the predicted class, repeat from 1 to total
number of training data points
 We have to calculate the distance between test data
and training data (each row). We will use the
“Euclidean distance” as our distance metric.
 We need to sort the distances in ascending order
based on distance values

2 019
 We will get top k rows (from the sorted array)
 Get the most frequent class of these rows

Year
 Return predicted class [11] [2][3]
Through this algorithm, we get Kth best
25
employee list.
Now the process of decision tree algorithm is Figure 3: Dataset 2

Global Journal of Computer Science and Technology ( D ) Volume XIX Issue IV Version I
given bellow. Explanation of the Dataset 2:
 Pick the best attribute/feature. The best attribute is The attributes of the DATASET 2 given in are-
one that best splits or separates the data. Experience, Cumulative Grade Point Average, Skill,
 Ask the relevant question. Publication and Training and Language
 Follow the answer path. Proficiency. the explanation of all the attributes of
 Go to step 1 until we arrive at the answer [12] Dataset 2 is bellowed.
According to the discussion, we can determine a) Experience
the summary of the methodology in such a process, According to the above-given scale, a fresher
which is defined as: candidate gets the minimum score, which is -1
A. Load all the application list according to our range of values. A candidate with
B. Receive working tools (W) name from the client experience of more than eight years gets the maximum
C. Receive experience range from the client. The range score, which is 1. Except for the score -0.75 for less than
should be (0<=N<=L); N= employee working two years of experience, in case of all other intervals, the
experience and L = clients experience requirement score changes with a step of 0.5.
D. Remove applicants list
when W =! Skill of applicant Table 1: [Experience]
E. Remove applicant list Number of years of
Score
if, (0<=N<=K) is not true experience
No experience (fresher) -1
F. Input the value of Marks and Potential from the client
Experience < 2 -0.75
Load the final list after removing
G. Give analytical exam and a video interview by the 2 < experience <= 4 -0.5
employee 4 < experience <= 6 0
H. Input the value of K, K= the number of employee the 6 < experience <= 8 0.5
clients need Experience > 8 1
I. Input the value of Marks and Potential from the client
b) Cumulative Grade Point Average
Use K nearest neighbor, decision tree, Scoring the CGPA is much simpler. As
Find the employee list [8] maximum CGPA possible is 4.00, we considered that to
be one under our scoring range, which is the maximum.
For accuracy, we tried a new dataset [5] [6]
Any other CGPA in the scale of 4 is convert into the
scale of 1.
For instance, a candidate with CGPA 4.00 gets
a maximum score of 1 added to his CGPA feature. If the
CGPA was 3.5 then the score would be,
Score = 3.5 / 4 = 0.875
So here, the CGPA 3.5 out of the scale of 4 has
got converted to 0.875, which is out of 1.

© 2019 Global Journals

 Employee Culling based on Online Work Assessment through Machine Learning Algorithm

c) Skill IV. Result Analysis

We have set the scoring policy depending on
the types of CVs that we have come across. Candidates According to DATASET 1, which is applied in
that have mentioned more than three skills, we fixed it to KNN and decision tree algorithm, we get that Shuvashish
score their skills feature to be maximum that means 1. [7th employee from DATASET 1] is the perfect employee.
But, those who have mentioned three or less, then there According to the accuracy from DATASET 2, we
we considered the degree of their skills in those areas or can deduce that SVM and Decision Tree algorithms
subjects. Which is bellowed. provide the best results. Although the Decision Tree
gives a higher percentage than the SVM algorithm.
Table 2: Skill
Table 3: Result Analysis
Good Very good Excellent
Decision Tree SVM Multi-Linear
2 019

0.1 0.2 0.3

85% 80% 72%
According to these criteria, suppose a
Year

candidate has mentioned about three skills MS Word, But we would recommend SVM. The reason is,
MS Excel, and PowerPoint. Consider the table below as that a decision tree starts the process of building a tree
26
an example, from scratch every time the algorithm is calling but with a
different root node and hence gives more volatile results
Global Journal of Computer Science and Technology ( D ) Volume XIX Issue IV Version I

Table 2.1 [Skill]

as well as being more prone to over-fitting as the
Skill Good Very good Excellent complexity of the dataset increases.
MS Word √
V. CONCLUSION
MS Excel √
PowerPoint √ Human resources management has a vision that
all the customers feel like a part of the community.
From the above skills and their degrees Therefore a company will always be met with prepared
mentioned by a certain candidate if we calculate the and helpful employees. HR management considered
score, their employees as important resources. On the contrary,
Score = MS word (Very good) + MS Excel today's innovation and development is machine learning.
(Very Good) + PowerPoint (Excellent) = 0.2 + 0.2 + This innovative subject is much more than we think,
0.3 = 0.7 which can develop anything as human wants. So in this
d) Publication and Training paper, for employees and companies betterment, we
In case of the features like training and create an internal and external bonding between human
publications, we have considered the number of training resource management and machine learning through the
programs a particular candidate has attended and the objective of the HR manager and machine learning
number of publications they have made. We focused algorithm.
mostly on the numbers rather than anything else. For
References Références Referencias
each of the publications that a candidate has made, 0.2
gets added to the publications feature. For example, 1. http://www.whatishumanresource.com/human-
someone with 4 publications get a score of 4 * 0.2 = resource-management
0.8. Just the same as that for each of the training 2. The Implementation of K-Nearest Neighbor
programs attended, a candidate gets same score 0.2. A Algorithm in Case-Based ReasoningModel for
candidate with five pieces of trainings attended gets a Forming Automatic Answer Identity and Searching
score of 5 * 0.2 = 1. Answer Similarity ofAlgorithm Case Yana Aditia
Gerhana1, Aldy Rialdy Atmadja2, Wildan Budiawan
e) Language Proficiency
Zulfikar3, Nurida Ashanti4 1,2,3,4Department of
For the enumeration of the feature English
Informatics UIN Sunan Gunung Djati Bandung Jl.
language proficiency, we considered the IELTS score to
AH Nasution No. 105, Bandung, West Java,
be the most standard scale to measure with an IELTS
Indonesia [email protected],
score of 7 or more will be taken as the maximum value 1
[email protected], [email protected],
for this feature. If it is less than 7, then the feature value
[email protected]
gets curbed according to the scale of 1.
3. Comparative Analysis of K-Nearest Neighbor and
For instance, if a candidate has an IELTS marks
Modified K-Nearest Neighbor Algorithm for Data
of seven then it adds 1 to the feature English
Classification
Proficiency, where one is the maximum. If there is some
4. The Amazing Ways How Unilever Uses Artificial
other candidate that has a score of 6.5 then the score is,
Intelligence to Recruit & Train Thousands of
Score = 6.5 / 7 =0.928; which is approximately one and
Employees[https://www.forbes.com/sites/bernardm
the score is good.
arr/2018/12/14/the-amazing-ways-how-unilever-
© 2019 Global Journals
 Employee Culling based on Online Work Assessment through Machine Learning Algorithm

uses-artificial-intelligence-to-recruit-train-thousands-
of-employees/#593b01ca6274]
5. Okfalisa Department of Informatic Engineering UIN
Sultan Syarif Kasim Riau Pekanbaru - Indonesia
[email protected]
6. An Adaptive k-Nearest Neighbor Algorithm
7. Shiliang Sun Rongqing Huang Department of
Computer Science and Technology, East China
Normal University 500 Dongchuan Road, Shanghai
200241, P.R. China [email protected],
[email protected]

2 019
8. A Quick Evidential Classification Algorithm Based
On K-Nearest Neighbor Rule Zhuang Wang, Wei-

Year
Dong Hu, Wen-Xian Yu Atr State Key Lab, National
Univ. of Defense Technology, Cbangsha, 410073,
China E-MAIL [email protected] 27
9. https://bizfluent.com/info-8212981-recruitment-

Global Journal of Computer Science and Technology ( D ) Volume XIX Issue IV Version I
process-hr.html
10. Uniliver application process [https://www.unilever.co
.za/careers/graduates/application-process/]
11. Introduction to K nearest neighbors: A powerful
machine learning algorithm(with implementation in
python & R) Tavish Srivastava
12. https://www.hackerearth.com/practice/machine-
learning/machine-learning-algorithms/ml-decision-
tree/tutorial/
13. Amazon scraps secret AI recruiting tool that showed
bias against women [https://www.reuters.com/articl
e/us-amazon-com-jobs-automation-insight/amazon-
scraps-secret-ai-recruiting-tool-that-showed-bias-
against-women-idUSKCN1MK08G]

© 2019 Global Journals