https://doi.org/10.31449/inf.v46i5.4079 Informatica 46 (2022) 123–128 123 
Evaluating Employee Performance with an Improved Clustering 
Algorithm 
Ci Fan 
Zhengzhou University of Industrial Technology, Zhengzhou, Henan 451150, China 
E-mail: f318to@163.com 
Keywords: clustering algorithm, performance appraisal, scoring method, appraisal indicator, density parameter 
Received: February 16, 2021 
This paper introduces performance appraisal management, analyzed its principles, proposes seven 
appraisal indicators, puts forward an improved K-means algorithm for the classification of appraisal 
results based on the K-means clustering algorithm and density parameters, and analyzes the performance 
of the method. The results of the UCI data set show that the accuracy rate of the improved K-means 
algorithm is significantly higher than that of the traditional K-means algorithm, and its highest accuracy 
rate is 91.27%. The classification of the appraisal results of 45 employees in Company A shows that the 
classification results obtained using the improved K-means algorithm are more in line with the actual 
situation and more reasonable and relevant than the traditional scoring method. The appraisal results of 
1064 technical employees show that most of the employees have a good performance. The results of the 
experiment verify the effectiveness of the improved K-means algorithm in performance appraisal 
management. The algorithm can be further promoted and applied in practice. 
Povzetek: Opisan je nov sistem za ocenjevanje kvalitete prispevkov sodelavcev s pomočjo gručenja.  
 
1 Introduction 
In order to make employees more motivated and 
responsible, it is a very important element to do a good 
performance appraisal management of employees [1]. 
With the development of society, people’s life is also 
increasingly moving towards informationization and 
intelligence, and enterprise management also applies more 
technological tools [2]; therefore, performance appraisal 
management through a more intelligent way is of great 
significance to improve employee performance and 
increase the sense of belonging [3]. Most of the current 
performance appraisal management has a strong 
subjective nature and relies more on the subjective 
judgment of the appraiser, which cannot achieve objective 
and fair appraisal of employees; therefore, with the 
development of information technology, there are more 
and more intelligent methods applied in the performance 
appraisal management of company employees. Yousif et 
al. [4] conducted a study on the performance evaluation of 
faculty members in Sudanese universities and developed 
appropriate criteria. Then, relevant data were obtained 
through expert assessment. Faculty members were 
comprehensively evaluated through hierarchical analysis 
and a technique for order preference by similarity to an 
ideal solution (TOPSIS). Ditzian et al. [5] evaluated three 
supervisors using the Performance Diagnostic Checklist–
Human Services (PDC-HS) and found that the PDC-HS-
indicated intervention (graphed feedback) was effective in 
improving performance. Bird et al. [6] considered that a 
performance management approach combining individual 
competencies with expertise could create significant 
values within highly educated work forces. Karpenko et 
al. [7] proposed an approach based on the overall key 
performance indicators (KPI) to analyze employee 
performance and found that through the use of KPIs, the 
process of operational activities can be better controlled 
and productivity can be improved. In this paper, an 
improved K-means algorithm was used to realize the 
performance assessment management of employees, and 
experimental analysis was carried out on Company A to 
verify the effectiveness of the method. The proposed 
method can realize a reliable performance management 
assessment, create a good working environment for 
employees, improve their motivation and promote better 
and faster development of the company. 
2 Employee performance appraisal 
management indicators 
Employee performance appraisal management is an 
important part of enterprise management [8], which is 
closely related to many tasks of enterprise management. 
For example, when conducting employee training, the 
results of the appraisal can reflect the technical level of 
employees, so training will be more targeted; when 
promoting positions, the results of the appraisal can help 
understand the ability, quality and attitude of employees; 
therefore, performance appraisal management has a 
certain effect on improving employees’ motivation and 
dedication [9]. With the development of the economy and 
the increase of the importance of performance appraisal 
[10], the amount of employee performance information 
has become larger and larger, and the traditional methods 
are increasingly unable to meet the current needs of 
company management [11]. There are more and more new 

124 Informatica 46 (2022) 123–128 C. Fan 
methods that are applied for analysis in performance 
appraisal. 
The performance appraisal management of employees 
generally follows the following principles. 
(1) Openness principle: the development, criteria, 
process and result of the appraisal should be open. 
(2) Objectivity principle: the criteria and process of the 
appraisal should be objective and avoid subjective 
emotions as much as possible. 
(3) Identity principle: the same standard and scale are 
used to appraise employees. 
(4) Institutional principle: performance appraisal 
management should be fixed as a system and strictly 
enforced. 
(5) Feedback principle: the results of the appraisal must 
be fed back to the employees and relevant 
departments. 
(6) Comprehensiveness principle: the content of the 
appraisal should be as comprehensive as possible. 
In order to conduct performance appraisal, the indicators 
for appraisal need to be developed first. Through literature 
review and expert consultation, seven indicators are 
identified in this paper, which is as follows. 
(1) Team spirit: able to obey the work schedule of the 
team. 
(2) Creative spirit: the ability to come up with innovative 
ideas and approaches when faced with problems. 
(3) Sense of responsibility: a high sense of responsibility 
to the company and work. 
(4) Cost consciousness: able to use company resources 
effectively and save costs in the workplace. 
(5) Working ability: Good grasp of relevant knowledge 
and skills, with high efficiency and high performance. 
(6) Communication skills: smooth communication with 
all levels of departments and colleagues without 
conflicts. 
(7) Moral quality: good conduct, decent behavior and 
good image. 
After evaluating the scores of the above seven indicators, 
this paper uses an improved clustering algorithm to 
process and classify the indicator evaluation results into 
four categories: excellent, good, qualified, and 
unsatisfactory. 
3 An improved clustering algorithm 
The principle of a clustering algorithm is to divide the 
huge amount of data into different clusters according to 
certain similarities, so that objects with high similarity can 
be in the same class [12]. The clustering algorithm can 
analyze data without knowing the size and distribution 
[13], so it has good processing effects on the uncertainty 
of data [14]. The K-means algorithm is a commonly used 
classification-based algorithm with very common 
applications in the classification of data [15] and performs 
well in fields of medical detection [16] and image 
segmentation [17]. Its calculation steps are as follows. 
(1) The number of clusters to be divided ( K ) is 
determined. According to the results of the 
performance assessment, K = 4 in this paper. 
(2) For the data set (x
1
, x
2
, x
3
, x
4
, x
5
, x
6
, x
7
), seven initial 
clustering centers, (𝐶 1
, 𝐶 2
, 𝐶 3
, 𝐶 4
, 𝐶 5
, 𝐶 6
, 𝐶 7
) , are 
randomly selected. The distance from every object to 
the center is calculated, and then every object is 
divided into the closest class. 
(3) The central point of every cluster is recalculated, and 
the formula is: 
𝐶 ′
𝑖 =
1
𝑛 𝑖 ∑ x
j x
j
∈C
i
, 
where i = 1,2,3,4,5,6,7 and 𝑛 𝑖 refers to the number 
of objects in class 𝐶 𝑖 . 
(4) If any 𝐶 𝑖 = 𝐶 ′
𝑖 holds, the algorithm converges, and if 
not, the algorithm returns to the second step. 
(5) The final clustering result is output. 
The result of the K-means algorithm is easy to 
understand and stable; however, its result has some 
dependence on the selection of the initial clustering 
centers. To solve this problem, this paper improves the K-
means algorithm by combining the density parameter of 
objects. 
The density parameter of an object refers to the 
distance between an objective and the nearest Minpts-th 
(constant) object [18] centered on that object, denoted by 
ε. The greater the value, the lower the density of the region 
in which the object is located. In this paper, the parameter 
is improved: a new m-dist value is put forward, which is 
the mean value of the distance between an object and the 
nearest Minpts-th (constant) object. The flow of the 
improved K-means algorithm is as follows. 
(1) For a given data set, all m-dist values are calculated 
based on the constant Minpts, and then the mean 
value, mean-dist, is calculated to obtain the set of 
density values, C. 
(2) The object with the lowest density in set C is used as 
the first initial cluster center. 
(3) The rest cluster centers are obtained in the same 
manner. Finally, seven initial clustering centers 
located in high-density areas are identified: 
(𝐶 1
, 𝐶 2
, 𝐶 3
, 𝐶 4
, 𝐶 5
, 𝐶 6
, 𝐶 7
). 
(4) Clustering is performed according to the K-means 
algorithm, and the result is output. 
4 Example analysis 
To verify the performance of the improved K-means 
algorithm, experiments were first conducted on the 
standard UCI data set [19], as shown in Table 1. 
The clustering analysis of the above data set was 
performed using the traditional K-means algorithm and 
the improved K-means algorithm, respectively, and the 
experimental results are shown in Figure 1. 
It was seen from Figure 1 that the accuracy of the 
traditional K-means algorithm ranged from 60% to 75%, 
with a minimum of 64.26% and a maximum of 74.51%, 
while the accuracy of the improved K-means algorithm 
was above 80%, with a minimum of 81.94% and a 
maximum of 91.27%, showing its reliability. The above 
results demonstrated that the accuracy of the algorithm 
was significantly enhanced after improvement; thus, it 

Evaluating employee performance with an improved clustering algorithm Informatica 46 (2022) 123–128 125 
would have better performance in performance appraisal 
management. 
Then, to study the application effect of the improved 
K-means algorithm in employee performance appraisal 
management, Company A was taken as an example. The 
positions in the company included technical positions, 
management positions and scientific positions, and 15 
employees were randomly selected from each of the three 
positions for performance management appraisal. The 
evaluators determined the score of every indicator 
according to the characteristics of different positions. The 
score of every position on every indicator is shown in 
Table 2. The total score of the appraisal was 70 points. 
The indicators were scored by the appraisers, and the 
scoring results are shown in Table 3. 
The assessment results were classified using the 
traditional scoring method and the improved K-means 
algorithm, respectively, and the classification criteria of 
the scoring method were: excellent for 45 points and 
above, good for 40-45 points, qualified for 35-40 points, 
and poor for 35 points and below. The final classification 
results of the two methods are shown in Table 4. 
It was seen from Table 4 that the results obtained by 
the two classification methods were not exactly the same, 
and there were obvious differences. The results of 
technical positions were used for specific analysis. Firstly, 
the scores of different indicators for employees in 
technical positions are shown in Table 5. 
It was seen from Table 4 that there were some 
differences between the assessment results of the two 
methods, and the data in Table 5 were combined to 
analyze the characteristics of different employees. The 
analysis results are as follows. 
(1) Employee 7: Table 5 shows that Employee 7 had a 
total score of 48 points, i.e., excellent in the scoring 
method, but in terms of the specific indicators, his 
score in “moral quality” was only four points, which 
was one of the lowest scores among the 15 employees 
who participated in the assessment, while the scores 
of the remaining six indicators were all high, 
Databa
ses 
Instan
ces 
Number 
of classes 
(k) 
Number 
of 
features 
(k) 
Size of 
classes 
Iris 150 3 4 505050 
Wine 178 3 13 597148 
Diabet
es 
768 2 8 268500 
Heartst
atlog 
270 2 13 150120 
Ionosp
here 
351 2 34 126225 
Table 1: Experimental data set. 
 
Figure 1Results of the experiment on the UCI data set.  
 Technical 
positions 
Management 
positions 
Scientific 
research 
positions 
Team spirit 12 10 12 
Spirit of 
innovation 
8 8 14 
Awareness of 
responsibility 
12 12 8 
Cost awareness 8 8 10 
Working ability 12 10 10 
Communication 
skills 
8 12 6 
Moral quality 10 10 10 
Total score 70 70 70 
Table 2: Score of different positions (unit: point).  
Nu
mb
er 
Te
a
m 
sp
iri
t 
Spir
it of 
inn
ova
tion 
Aw
are
nes
s of 
resp
onsi
bilit
y 
Cos
t 
awa
ren
ess 
Wo
rkin
g 
abil
ity 
Co
mm
uni
cati
on 
skil
ls 
Mo
ral 
qua
lity 
To
tal 
sc
ore 
1 
9 6 10 8 10 7 8 58 
2 7 7 7 7 7 7 7 49 
3 7 5 7 6 3 5 7 40 
4 5 6 7 8 9 7 8 50 
5 6 4 3 4 5 4 5 31 
6 2 7 7 5 7 5 8 41 
7 8 7 7 5 9 8 4 48 
...... ....
.. 
...... ...... ...... ...... ...... ...... .....
. 
43 7 7 5 5 5 4 6 39 
44 8 9 5 7 7 5 5 46 
45 7 8 5 5 5 4 4 38 
Table 3: Scores of every employee on different indicators. 

126 Informatica 46 (2022) 123–128 C. Fan 
indicating that his performance in moral quality was 
poor. Therefore, it was more reasonable to classify 
him as “qualified” in the improved K-means 
algorithm. 
(2) Employee 3: Table 5 shows that the total score of 
Employee 3 was 40 points, i.e., good in the scoring 
method, but in terms of the specific indicators, his 
score in the “work ability” was only three points, 
indicating that his work ability was poor so that he 
was inappropriate for a technical position. Therefore, 
it was more reasonable to classify him as “qualified” 
in the improved K-means algorithm. 
(3) Employee 6: Table 5 shows that Employee 6 had a 
total score of 41 points, but his score on “team spirit” 
was only two points, which indicated that his 
performance in team spirit was extremely poor and he 
might have disobedience and inability to cooperate 
well. Therefore, it was more reasonable to put him in 
the “poor” category in the improved K-means 
algorithm. 
(4) Employee 15: Table 5 shows that the total score of 
Employee 15 was 39 points, i.e., qualified in the 
scoring method, but he was classified as good in the 
improved K-means algorithm. Specifically, the 
performance of Employee 15 was balanced in all 
indicators, all above five points, so it was more 
reasonable to classify him as “good”. 
The comparison and analysis of the appraisal results found 
that compared with the traditional scoring method, the 
improved K-means method obtained more practical 
results and provided a more comprehensive and detailed 
appraisal of employees, which can provide more reliable 
and powerful support for the performance appraisal 
management of employees. 
The performance appraisal of 1064 technical 
employees in Company A was conducted using the 
improved K-means method, and the results are shown in 
Figure 2. 
  The 
scoring 
method 
The improved 
K-means 
algorithm 
Technica
l 
positions 
Excellent 1, 2, 4, 7, 
9. 
1, 2, 4, 9 
Good 3, 6, 8, 12 8, 12, 15 
Qualified 14, 15 3, 7, 14 
Poor 5, 10, 11, 
13 
5, 6, 10, 13 
Manage
ment 
positions 
Excellent 16, 25, 28, 
29 
16, 25, 28 
Good 19, 21, 24, 
30 
18, 19, 24, 30 
Qualified 17, 18, 20, 
26, 27 
20, 21, 22, 26, 
27, 29 
Poor 22, 23 17, 23 
Scientific 
research 
positions 
Excellent 31, 32, 37, 
42, 44 
31, 32, 44 
Good 34, 35, 38 34, 38 
Qualified 33, 40, 41, 
43, 45 
35, 37. 
Poor 36, 39 42, 36, 39 
Table 2Comparison of classification results. 
Nu
mb
er 
Tea
m 
spir
it 
Spir
it of 
inn
ova
tion 
Aw
are
nes
s of 
resp
onsi
bilit
y 
Cos
t 
awa
ren
ess 
Wo
rkin
g 
abil
ity 
Co
mm
uni
cati
on 
skil
ls 
Mo
ral 
qua
lity 
To
tal 
sc
or
e 
1 9 6 10 8 10 7 8 58 
2 7 7 7 7 7 7 7 49 
3 7 5 7 6 3 5 7 40 
4 5 6 7 8 9 7 8 50 
5 6 4 3 4 5 4 5 31 
6 2 7 7 5 7 5 8 41 
7 8 7 7 5 9 8 4 48 
8 6 5 6 5 7 5 7 41 
9 6 6 8 6 7 6 8 47 
10 7 4 2 4 7 4 6 34 
11 5 4 5 4 5 4 4 31 
12 7 6 7 6 7 5 6 44 
13 4 4 5 4 5 4 4 30 
14 6 5 3 6 8 5 6 39 
15 6 5 6 6 6 5 5 39 
Table 1Score results of each index for technical positions.  
 
Figure 2Technical position assessment results. 
 

Evaluating employee performance with an improved clustering algorithm Informatica 46 (2022) 123–128 127 
It was seen from Figure 2 that the number of technical 
employees who were assessed as “good” was the largest, 
reaching 38%, followed by “excellent” employees, 
reaching 32%, and “qualified” employees, reaching 27%, 
and the number of employees who were assessed as “poor” 
was the lowest, 3%. According to the above results, most 
technical employees had a good performance, but there are 
also a small number of employees with poor performance. 
The analysis of the specific indicators suggested that the 
employees with poor performance either had an obvious 
lack of ability in a certain area or had poor overall capacity. 
For poorly performed employees, the company needs to 
carefully consider their salaries and promotions; for well-
performed employees, the company should make 
appropriate encouragement to mobilize their enthusiasm. 
The experimental analysis found that the improved K-
means algorithm had favorable performance in 
performance appraisal management, which is conducive 
to improving the level of company management, 
promoting the efficiency and quality of management, 
establishing a reasonable human resource incentive 
mechanism, helping employees to discover their 
shortcomings, improve their working ability, and drive 
team development to promote company progress. 
5 Conclusion 
Based on the clustering algorithm, this paper studied the 
method of company employee performance appraisal 
management. An improved K-means algorithm was 
designed, and its effectiveness was verified through 
experiments on the UCI data set. Then, the actual 
employee performance data were classified. The 
comparison between the improved algorithm and the 
traditional scoring method showed that the improved 
algorithm had higher rationality and could be further 
promoted and applied in practice. This work makes some 
contributions to improve the performance appraisal 
management effect and promote the common progress of 
employees and companies. 
References 
[1] Moradi T, Mehraban M A, Moeini M (2017). 
Comparison of the Perceptions of Managers and 
Nursing Staff Toward Performance Appraisal. 
Iranian journal of nursing and midwifery research, 
22, pp. 128. 
[2] Najjar A, Amro B, Macedo M (2021). An intelligent 
decision support system for recruitment: resumes 
screening and applicants ranking. Informatica, 45, 
pp. 617-623. 
[3] Gomez-Mejia LR (2016). Increasing Productivity: 
Performance Appraisal and Reward Systems. 
Personnel Review, 2016, 19, pp. 21-26. 
https://doi.org/10.1108/00483489010138759 
[4] Yousif M K, Shaout A (2016). Fuzzy logic 
computational model for performance evaluation of 
Sudanese Universities and academic staff. Journal of 
King Saud University - Computer and Information 
Sciences, 30, pp. 80-119.  
https://doi.org/10.1016/j.jksuci.2016.08.002 
[5] Ditzian K, Wilder D A, King A, Tanz J (2015). An 
evaluation of the Performance Diagnostic Checklist-
Human Services to assess an employee performance 
problem in a center-based autism treatment facility. 
Journal of Applied Behavior Analysis, 48, pp. 199.
https://doi.org/10.1002/jaba.171 
[6] Bird H (2015). Appraising clever people: lessons 
from introducing performance reviews for 
academics in a UK University. Industrial & 
Commercial Training, 47, pp. 81-85. 
https://doi.org/10.1108/ICT-10-2014-0069 
[7] Karpenko A, Karpenko N, Shudrik O (2020). 
Development and implementation of a strategic 
personnel management system according to goals 
based on key performance indicators. Management 
and Entrepreneurship Trends of Development, 2, pp. 
22-35.  
https://doi.org/10.26661/2522-1566/2020-2/12-02 
[8] Uysal G (2016). Cognitive Placement Theory for 
Performance Appraisal: Talent Management and 
Individual Performance. Journal of Modern 
Accounting and Auditing, 012, pp. 60-63.  
https://doi.org/10.17265/1548-6583/2016.01.005 
[9] Nair M S, Salleh R (2015). Linking Performance 
Appraisal Justice, Trust, and Employee Engagement: 
A Conceptual Framework. Procedia - Social and 
Behavioral Sciences, 211, pp. 1155-1162.  
https://doi.org/10.1016/j.sbspro.2015.11.154 
[10] Andersson G (2016). Data Recording in 
Performance Management: Trouble With the Logics. 
American Journal of Evaluation, 38, pp. 190-204. 
https://doi.org/10.1177/1098214016681510 
[11] Lonsdale A (2016). Performance Appraisal, 
Performance Management and Quality in Higher 
Education: Contradictions, Issues and Guiding 
Principles for the Future. Australian Journal of 
Education, 42, pp. 303-320.  
https://doi.org/10.1177/000494419804200307 
[12] Lv Y, Ma T, Tang M, Cao J, Tian Y, Al-Dhelaan A, 
Al-Rodhaan M (2016). An efficient and scalable 
density-based clustering algorithm for datasets with 
complex structures. Neurocomputing, 171, pp. 9-22. 
https://doi.org/10.1016/j.neucom.2015.05.109 
[13] Kumar Y, Sahoo G (2014). A chaotic charged system 
search approach for data clustering. Informatica, 38, 
pp. 249-261. 
[14] Guo Y, Sengur A (2015). NCM: Neutrosophic c-
means clustering algorithm. Pattern Recognition, 48, 
pp. 2710-2724.  
https://doi.org/10.1016/j.patcog.2015.02.018 
[15] Dhanachandra N, Manglem K, Chanu Y J (2015). 
Image Segmentation Using K -means Clustering 
Algorithm and Subtractive Clustering Algorithm. 
Procedia Computer Science, 54, pp. 764-771. 
[16] Abaker AA, Saeed FA (2021). A Comparative 
analysis of machine learning algorithms to build a 
predictive model for detecting diabetes 
complications. Informatica, 45, pp. 117-125. 

128 Informatica 46 (2022) 123–128 C. Fan 
[17] Jiang HN (2018). Defect features recognition in 3d 
industrial ct images. Informatica, 42, pp. 477-482. 
[18] Yuan F, Zhou ZY, Song X (2007). K-means 
Clustering Algorithm with Meliorated Initial Center. 
Computer Engineering, 33, pp. 65-66.  
https://doi.org/10.3969/j.issn.1000-
3428.2007.03.024 
[19] Zhang Y, Zhang P (2015). Machine training and 
parameter settings with social emotional 
optimization algorithm for support vector machine. 
Pattern Recognition Letters, 54, pp. 36-42.  
https://doi.org/10.1016/j.patrec.2014.11.011