The Application of Data Mining Techniques in Employee Performance

Assessment

ZHAO ZHENG

Lyceum of the Philippines University Manila Campus,

Manila 1002,

PHILIPPINES

Abstract: - Employee performance assessment is a powerful standard for measuring talent, and many

able to measure employee performance well, the accuracy of the decision tree algorithm is 93.2%, and the

accuracy of the improved algorithm is 95.3%, so the improved algorithm is 39 ms shorter than the traditional

algorithm in building the decision tree, and the algorithm accuracy is 2.1% higher. This shows that the

improved decision tree algorithm of data mining technology can improve the precision and accuracy of

employee performance evaluation.

Key-Words: - Employee Performance Measurement; Data Mining; Decision Tree Algorithm; Improvement;

Improved algorithm; Algorithm comparison; Accuracy.

Received: October 25, 2022. Revised: September 9, 2023. Accepted: November 7, 2023. Published: December 4, 2023.

an important means, [1]. The rapid development of

information data makes the information data in

many fields explode, how to make good use of this

information has become a problem worth thinking

about, [2]. Although information mining technology

can complete the processing and analysis of data

through many algorithms, the current existence of

many data mining algorithms is not perfect, [3].

Currently, there are many performance evaluation

methods, but most of them can only obtain some

current data mining algorithms are improved, and a

new data mining improvement decision tree

algorithm employee performance evaluation model

is constructed to deeply mine employee

performance data. This can greatly compensate for

the shortcomings of traditional methods. This

experiment is divided into four parts, the first part is

the elaboration of domestic and international

scientific research results; the second part mainly

introduces the basic concepts of the research

problem and the construction process of the

algorithm model; the third part proves the feasibility

of the algorithm through experiments; and the fourth

part summarizes the whole research process.

2 Related Works

With the continuous development of science and

technology, data mining technology is more and

more widely used in various fields. Among them,

the decision tree algorithm (DTA), as an excellent

confidence level of the DTA in the hypothetical

dataset. The experimental results show that the

combined algorithm avoids the problem of

overfitting, and the generalization performance is

significantly improved. [6], proposed an improved

DTA to evaluate the skin-wise sensitivity in the

cosmetic safety index. The experimental results

showed that the algorithm was more efficient than

the traditional assessment model, with better clarity

classification and developed a set of flowcharts. The

experimental results showed that the incorporation

of the algorithm greatly simplified the students'

steps and reduced the difficulty of the production

process, [7]. [8] proposed a modeling framework

based on gradient-enhanced decision trees to better

detect the residual capacity of lithium-ion batteries

for electric vehicles. The framework experiments

with an accurate assessment of the battery's health

capacity by extracting effective indicators from the

efficiency and higher validity of the remaining

battery life than the traditional method.

measurement is widely used to assess the

performance of employees and their contribution to

demonstrated by employees in their work, [9] To

expand the performance assessment of professionals

in academic libraries, a structural equation model of

performance appraisal was introduced to 339

perceptions of a high-performance work system,

524 employees of three Greek manufacturing

companies were subjected to an experimental

performance measurement over a period of three

months. The results of the experiment showed that

employees' job satisfaction was positively related to

job stress, role ambiguity, and role conflict. Thus

this performance measure provides ideas for better

regulation of high motivation of employees at work.

1200-point questionnaire from the public and

private sectors of Pakistan. The findings indicated

that personal and work-related text message use can

improve employee performance through knowledge

exchange. [12], investigated the mechanism of

employee non-ethical organizational behavior on

employees and supervisors, 304 employees and 94

supervisors in several manufacturing industries in

China were used as research subjects to test the

hypotheses using a hierarchical linear model. The

organizational behavior is positively related to

supervisors' performance appraisal, and at the same

time supervisors' employee non-ethical

organizational behavior indirectly affects

employees' performance appraisal.

focusing on the field of employee performance

measurement, but there are still some challenges and

problems. For example, the traditional assessment

model is based on qualitative evaluation, lacks

of employee performance evaluation, to provide a

theoretical basis and new ideas for the research in

this field.

Mining Techniques in Employee

Performance Measurement

tree improvement algorithm employee performance

measurement model is obtained.

Performance assessment is an essential task in the

enterprise, which can test and evaluate the salary

status of employees, so the performance assessment

system needs to have several characteristics. The

have a complete set of systematic methods. At the

same time performance assessment system must be

the need to have clarity of purpose to ensure the

reasonable operation of employee performance

assessment, [13]. Therefore, using data mining

through the processed data, and then data

conversion to achieve the mining of data, and finally

the mined data through the evaluation of patterns to

achieve the discernment of useful knowledge. The

function of data mining is to utilize the initial

characteristics of the data, including data noise

incompleteness, ambiguity, randomness, etc.

Meanwhile, the content of data mining can be some

semi-structured forms of data such as text and

Fig. 2: Cross-mining in different fields of data

different fields through machine learning, statistics,

and information science can achieve the analysis

and utilization of data when using data mining, and

database technology is used to mine data through

real-world problems. In general, the main function

of data mining is to characterize the data ensemble

of the object of study, and through the hidden

relationships and laws of these data predict the

the variables; cluster analysis function is to be able

to analyze the subset of a variety of data to cluster;

descriptive concepts are to conceptually describe

some things, and at the same time is divided into

characteristic description and distinguishing

description; bias detection refers to the existence of

the database to data present in the database; bias

detection refers to the detection of the data present

in the database by measuring its bias value and true

Fig. 3: Analysis of data mining processing structure

system includes the storage of data and the

pre-processed before the next operation. Data

mining processing is mainly for the data objects that

need to be processed, using data algorithms to mine

the relevant data, [15].

The commonly used algorithmic models for data

mining are the Decision Tree Algorithm, Bayesian

Classification Algorithm, Support Vector Machine

Algorithm, Genetic Classification Algorithm,

algorithm can classify a large amount of data with a

purpose, mining useful hidden data information, and

can be used for data prediction models, not only

does it have a high efficiency at the same time for

the construction of the algorithm is relatively simple,

so it applies to the construction of the algorithm

model of data mining, [16]. Decision Tree

Algorithm is a predictive algorithm model that can

be learned through the data training set as a

decision tree. The main flow of the decision tree

algorithm is shown in Figure 4.

of reality on the decision-making nodes to create,

and then identify the data samples belonging to the

same type of samples are returned to the leaf nodes,

information entropy can judge the information of

the node through the construction of the DTA to

complete the analysis. Let the data set of the sample

be

The sample is

, then the correct

information entropy formula is shown in Equation

(1), [17].

sample set;

indicates the number of samples of

the set

, if the type of occurrence of no sample

Equation (2).

weights of the subset of the dataset

of information as shown in Equation (3), [18].

this point can be expressed by Equation (4).

size of information gain efficiency,

information entropy, able to obtain different

function datasets through the unit cost. SLIQ

algorithm is an algorithmic model for high-speed

decision-making in decision tree algorithms. SLIQ

algorithm can replace the information entropy of the

among the various data sets, so at this time, the

denotes the nodes, and

denotes the quantity at

be interfered in the classification of continuous

attributes. The relative frequency of the fuzzy leaf

nodes in this case is shown in Equation (7).

in the non-decision algorithm and

represents

the classification of fuzzy attributes. Therefore, the

Equation (8).

Equation (8) for the fuzzy attribute

first

definition. In this case, the definition

formula for the fuzzy attribute

Equation (10).

(10)

fuzzy information entropy as the information

entropy fuzzy attribute for the current classification,

by using filters that increase the level of filtration

when classifying the information of the data, and

then by setting the real data parameter

to

formula for normalization is shown in Equation

(11).

and the defined uncertainty for the value of the

to-be-selected attribute

is shown in Equation (12).

the order of arrangement

uncertainty for non-leaf nodes is defined as shown

in Equation (13).

algorithm parameters.

From the construction process of the algorithm,

construction process of the decision tree algorithm

using ID3 is relatively complex, during which there

are a large number of repetitive operations, and at

the same time, due to the algorithm there are several

[20]. The decision tree model is usually constructed

as shown in Figure 5, taking weather data as an

example.

Fig. 5: Decision tree model

is on the high side, and then through the decision

judgment on the temperature and humidity and wind

whether the weather belongs to the sunny or rainy

days, while the cloudy days are listed separately,

which is the general decision process of the decision

tree.

algorithm process is, first of all, calculating sample

data for all attributes of the decision-making

calculations that appear similar or equal, at this

point in the calculation can no longer be decided to

split the attribute, it is necessary to another

calculated attribute gain and select the largest value

of which is the split attribute value. At this point the

decision function of the sample data belongs to the

same category of sample attributes and can be used

as a decision function of the leaf nodes, then the

decision-making algorithm to improve the algorithm

is completed. Because of the size of the true value

the classification, it is necessary to switch the

maximum value in the DTA and the minimum value

of the fuzzy information entropy. Therefore, the data

mining process for employee performance is shown

in Fig. 6, [21].

employee is collected such as name, work number,

department, etc. Secondly, the measured attributes

of the employee performance information are

divided into statistics, and then the raw data are

processed such as data conversion, data integration,

meets the predetermined needs and outputs the data

if it meets the requirements, and re-assesses and

calculates if it does not meet the requirements.

experimental data of a company is firstly selected as

the performance evaluation data, and the

performance information of employees is collected

during the data collection. Due to the large number

of collected data attributes and quantities, it is

necessary to pre-process the data for some data

data integration, screening, conversion, and so on.

Finally, a decision tree is constructed by improving

the decision tree algorithm to evaluate the

performance assessment data, [22].

. Then the information entropy of each performance

data is calculated by Equation (1), and different

difficult to speak about the progressive nature of

data changes reflected, for example, when the data

information and the grade division are only one

point different, 59 points will be classified as D, 60

points will be classified as C, the results of this

assessment is unfair. Calculate the degree of

membership for each level of performance, and

fuzzify the data to weaken the unfairness brought by

the data, thus achieving a transition from

In measuring employee performance, performance

is measured using the hold-up method. The data is

randomly divided into a test set and a training set.

a decision tree algorithm and one is a decision tree

improvement algorithm. The test results obtained

using number sets with different sample sizes are

shown in Figure 7.

number of samples reaches the maximum value the

decision tree building time reaches the maximum

value. At this time, the building time of the decision

tree algorithm is 229 ms, and the building time of

the improved decision tree algorithm is 190 ms,

with a difference of 39 ms between the two

maximum values. When comparing the accuracy of

the algorithms, it is found that the accuracy of the

algorithms increases when the samples increase.

time of the DTA, and the accuracy of the algorithm

is 2.1% higher. The fuzzy information of the

decision tree is compared with the data as shown in

Table 1.

as realism and skill level can react to the fuzzy

information of the current decision-making

algorithm, when the skill level and the degree of

enthusiasm for the work are A level, the assessment

results obtained are also A level, and the assessed

whose truthfulness is less than 100%, there are two

assessment results of grade A, one assessment result

of grade B, and one assessment result of grade C.

Table 1. Comparison of fuzzy information in

improved decision-making algorithm. The

evaluation results can reflect the decision-making

ability of fuzzy information of the current improved