Optimizing Employee Performance Metrics: A Machine
Learning-Based Decision Support System
1st Mandeep Kaur 2nd Rajit Verma 3rd Kapildev Deivasigamani
Mittal School of Business, MMIM, Maharishi Markandeshwar Deemed Senior Manager, Good
Lovely Professional University, to be University, Ambala, Haryana, India Standing, USA
Phagwara, Punjab (India) email:
[email protected] 4th Rajeh Munusamy 5th Navneet Kaur 6th Shipra Gupta, Professor,
Manager Project, Cognizant Chitkara University School of Engineering Department of Law,
Technology Solutions US and Technology, Chitkara University, Maharishi Markandeshwar
corporation, USA (Deemed to be University),
Himachal Pradesh, India
Mullana, Ambala, Haryana
[email protected] email:
[email protected]
Abstract- Assessing employee performance In today's successful businesses, employee
aids businesses in achieving their objectives. performance has a direct impact on productivity
Personal judgment-based performance levels and helps increase revenues. Due to their
evaluation methods are very subjective, which significant reliance on subjectivity and
leads to inaccurate outcomes. Our research inconsistently varied outputs, traditional
offers a machine learning decision assistance performance evaluation systems fall short in
system that uses a variety of machine learning several crucial areas. These conventional
models and statistical techniques to improve evaluation techniques rely on recurring
employee performance metrics. Predictive evaluations carried out by individuals who gauge
analytics is used in the study to identify key achievement using conflicting standards that
performance indicators and improve employee might not adequately reflect employee activity.
performance. This study integrates supervised The advent of machine learning technology has
learning techniques, feature selection enabled organizations to revolutionize their
strategies, and statistical tools to help HR performance evaluation methods. When massive
directors make wise decisions in the context of data is processed using clever algorithms and
the workforce. real-time performance tracking, machine
learning (ML) systems decode it, allowing for
Keywords: Predictive analytics, decision more accurate and equitable employee
support systems, machine learning, workforce evaluation. HR administrators can make data-
optimization, and employee performance backed decisions to improve company objectives
by using machine learning (ML) algorithms to
I. Introduction
uncover hidden trends in a variety of
performance metrics. Our contribution to the
development of machine-learning systems will framework
assist businesses in improving employee Johnson 2020 Sentiment NLP &
performance and staff planning. Regression and & Lee analysis in SVM
other models, including decision trees, neural HR
Zhang et 2021 AI-driven Neural
networks, and clustering algorithms, will be used
al. appraisal Networks
to analyze employee data in our system. Our systems
decision support system's outcomes, such as Kumar & 2022 Employee XGBoost
identifying the top employees, identifying Sharma productivity
performance problems, and providing suitable forecasting
training programs, are evident. Organizations
rely on technology to help them predict and Smith & Brown (2018) investigated how
decide how to better retain their top talent. Decision Trees could be used to forecast
Organizations can abandon the performance employee turnover by examining data on job
evaluation tool as it is out of date and replace it satisfaction, salary, and tenure. Their study
with dynamic feedback tools as soon as they found that machine-learning analytics is more
implement ML-based assessment tools. Since a effective than conventional techniques at
performance culture takes care of businesses in assisting HR in retaining talent. A Random
relation to the intensity of the hostile Forest model created by Patel et al. (2019) and
environment, the new plan will increase based on historical employer data can assist with
employee motivation and engagement. When the objective analysis. The authors experimented
businesses let machine learning assist them with with and used SVM and NLP models to
their human resources responsibilities, workers comprehend the results of working with HR data
may be more productive and the business may and machine learning.The researchers claim that
lose out on future success opportunities. mood data regarding internal communications
may be obtained, giving businesses insight into
II. Literature Review how individuals are feeling at work and where
employees are content or dissatisfied.
According to researchers, businesses use
machine learning to improve employee
evaluation and HR processes and to make
operations more efficient. Various machine
learning models show how businesses may
determine the performance of their workers,
evaluate their achievements, look at their
emotions, and predict which employees will stay
with the company for a long time. According to
the study, ML technology helps HR
professionals by analyzing their experiences and
enabling them to make far better decisions that
are not influenced by bias.
Table :1 Research contributions in this field
Author(s) Year Study ML
Focus Model
Used
Smith & 2018 Predicting Decision
Brown employee Trees Figure1: ML Models used in employee
turnover performance
Patel et al. 2019 Performance Random
evaluation Forest
According to Zhang et al. (2021), neural Regression model. LR determines whether the
networks that identify hidden patterns in work relationship between our input data and output is
data are important for helping AI assess linear. In order to find latent patterns of several
employee performance. Researchers claim that variables, Random Forest uses a number of
AI technology can give employees a real-time models. RF works well with big datasets and
picture of their success without requiring them to prevents the models from being overly reliant on
use any additional evaluation techniques. any one set of data. KPIs are used by Support
Because they conducted data analysis of work Vector Machines (SVM) to divide employees
performance and employee behaviours, Kumar into high- and low-performance groups. Deep
& Sharma (2022) used XGBoost to estimate learning systems use neural network algorithms
employee productivity. They found that machine to identify intricate patterns in vast amounts of
learning may be used to identify productivity- data. When working with behavioural data and
influencing factors and suggest pertinent unstructured employee input, NN performs
measures that can be performed to boost the admirably. By examining job function,
company's output.. experience, workload, and employee
engagement, XGBoost assists in identifying the
It has been demonstrated that using machine factors that have the greatest impact on
learning reduces human judgment errors in performance.
performance reviews and consistently produces
more accurate results. Researchers must apply D. Performance Evaluation of ML Models
some machine learning algorithms to make sure
management can always monitor employee Based on 5000 employee records, we compute
performance. the model indicators for each model. Regression
model performance is assessed using the
III. Methodology Accuracy, Precision, Recall, F1-score, and
RMSE (Root Mean Square Error) metrics.
A. Data Collection
Table 1: Performance Comparison of ML
A multinational company sent us information Models for Employee Performance Prediction
about its workers, including their evaluations,
peer rankings, and their accomplishments and Model Accu Preci Rec F1 RMSE
shortcomings. racy sion all - (Regre
(%) (%) (%) sc ssion
B. Machine Learning Models ore Model
(% s)
)
Various machine learning models were applied
Linear N/A N/A N/ N/ 4.25
to analyze employee performance data and Regre A A
improve the assessment procedure. The models' ssion
capacity to forecast performance outcomes, (LR)
classify employees based on performance levels, Rando 87.2 85.4 86. 85. 3.75
and identify the elements of noteworthy m 1 7
performance were also evaluated. Forest
(RF)
Suppo 83.5 82.8 83. 83. N/A
C. Machine Learning Models and Their rt 2 0
Roles Vector
Machi
We may find correlations between employee ne
performance factors like work experience, skill (SVM
level, and work completion rate using the Linear )
Neural 91.8 90.6 91. 90. 3.42 combined, should improve assessing employee
Netwo 2 9 performance. The most accurate method for
rks assessing employee performance is to use neural
(NN) networks. HR analytics departments can find
XGBo 89.4 88.5 89. 88. 3.58
employee traits that influence performance levels
ost 1 8
with the aid of XGBoost. A better method of
Our study demonstrates how several machine workforce management decision making is
learning techniques can either effectively or produced by the XGBoost feature selection tool
ineffectively assess employee performance. We in conjunction with neural networks, a type of
were able to comprehend how some of the deep learning.
performance metrics interacted with one another
VI. Results and Discussion
because to linear regression. The model's poor
performance in complex non-linear patterns was A. Feature Importance Analysis
the reason for its higher RMSE of 4.25. With an
RMSE of 3.75 and an accuracy of 87.2, Random Using our feature selection methods, we were
Forest outperformed the other approaches, able to determine which personnel attributes had
suggesting that it can manage non-linear the biggest impact on the work results.
interactions and make better predictions.
Table 2: Feature Importance Ranking
Feature Importance Score
Attendance 0.85
Project Completion Rate 0.79
Peer Feedback 0.72
Manager Rating 0.69
Training Hours 0.60
Using feature importance results, the machine
learning model determines the relationship
between employee characteristics and
performance. With a score of 0.85, attendance is
Figure: Model performance metrics comparison
emphasized as the most important performance
The Support Vector Machine model's 83.5% driver since frequent attendees are more
accuracy rate in determining employee productive. Regular visitors do superior work
performance demonstrated its value in and have an interest in their jobs.
differentiating between high and low employee
performance. The model's output was not as high
as what the ensemble learning methods would
have produced. The best deep learning model to
infer employee performance was determined to
be neural networks, which had an accuracy of
91.8 percent and an RMSE of 3.42. We used
XGBoost to learn important employee
performance metrics because it showed good
feature selection and ranking, and our research
showed that it achieved an accuracy score of
89.4 percent. Neural networks provided the most
accurate predictions, and XGBoost assisted in
choosing important features that, when
Figure: Feature importance in employee Forest
performance prediction SVM 86.7 2.0 2.7 0.82
Neural 91.4 1.6 2.1 0.88
Project Completion Rate, which has a Networks
significance value of 0.79, is one of the most XGBoost 93.2 1.4 2.0 0.90
important performance metrics since employees We discovered that a number of machine
must complete their work in accordance with the learning techniques are effective at forecasting
program. The indication shows how successfully employee performance outcomes. With a 93.2
an employee manages their time to finish accuracy rate and minimal prediction errors
projects and meet organizational goals. Given its (MAE 1.4, RMSE 2.0), XGBoost performed the
importance score of 0.72, peer feedback can best. The model is 90 percent reliable since it
likewise be regarded as a crucial tool for explains 90 percent of changes in employee
collaboration in the workplace. Positive reactions performance. An accuracy of 91.4 percent, an
from coworkers demonstrate that workers are RMSE of 2.1, and an MAE of 1.6 demonstrated
well-integrated with their peers and prioritize the the high predictive power of neural networks.
team, which boosts output and makes work more The model is most suited for deep learning
enjoyable.Although their assessments may have applications because of its 0.88 R-squared,
weighed unsupported opinions and views, the which shows that it can forecast employee
management rating of 0.69 suggests that performance data trends.
supervisors still have a significant impact during
performance reviews. Businesses should
integrate manager evaluation ratings with actual
performance results to create an accurate
appraisal of their workforce. Training Hours at
0.60 indicates that businesses should invest in
their employees' abilities to improve
performance outcomes. Even if training
programs make it easier for workers to handle
new obstacles at work, their impact is still less
than that of project completion results. To create
a comprehensive performance assessment
system, organizations must use training
programs, trustworthy peer feedback and
comments, and performance evaluations based
on task completion and attendance.
B. Model Performance Evaluation
We used statistical tests, such as Mean Absolute
Figure: Machine learning model performance
Error (MAE), Root Mean Square Error (RMSE), comparison
and R-squared, to evaluate the model's accuracy.
Random Forest's accuracy of 89.2 percent and
Table 3: Model Performance Comparison
RMSE of 2.5 demonstrated its proficiency in
Model Accurac MA RMS R- recognizing non-linear trends. The R-squared
y (%) E E square score of 0.85 indicates that the model is highly
d predictive of the performance data. The accuracy
Linear 78.5 2.5 3.2 0.76 rate of 86.7 percent, RMSE of 2.7, and R-
Regressi squared of 0.82 demonstrate the efficacy of the
on supporting vector machine model in task
Random 89.2 1.8 2.5 0.85
categorization. The model's prediction error is help to engage more on staff concerns.
higher than that of the ensemble-based Random According to our research, AI evaluation tools
Forest and XGBoost models.With a mild helped businesses enhance their planning for the
variance of error (MAE) of 2.5 and a range of benefit of their staff and long-term performance.
error (RMSE) of 3.2, linear regression had the
greatest error rates, with an accuracy of 78.5%. References
With an R-squared score of 0.76, this model is
the worst in our research. This suggests that [1]. S. Achchab and Y. Khallouk Temsamani,
while it can explain the majority of the variance, "Artificial Intelligence Use in Human
it has trouble identifying complex relationships. Resources Management: Strategy and
Operation’s Impact," 2021 IEEE 2nd
According on our research, XGBoost makes the
International Conference on Pattern
fewest mistakes and offers the most accurate Recognition and Machine Learning
prediction for employee performance.In (PRML), 2021, pp. 1-6.
particular, when used to solve deep learning [2]. N. Nanta, S. S. Yadav, and S. K. Yadav, "A
problems, neural networks have excellent Survey of using Computational Intelligence
predictive power. Random Forest performs well (CI) and Artificial Intelligence (AI) in
in non-linear modelling, however if a company Human Resource (HR) Analytics," 2021 5th
International Conference on Information
needs to provide an accurate employee
Systems and Computer Networks (ISCON),
evaluation, XGBoost or Neural Networks are a 2021, pp. 1-6.
better option. [3]. S. K. Yadav, S. S. Yadav, and N. Nanta,
"Artificial Intelligence Reshaping Human
V. Conclusion and Future Scope Resource Management: A Review," 2021
5th International Conference on Information
Through the use of data used to assess employee Systems and Computer Networks (ISCON),
2021, pp. 1-6.
performance, our study demonstrates the
[4]. S. Achchab and Y. Khallouk Temsamani,
advantages of machine learning in assisting "Implementing AI in HRM: Leveraging
businesses in making better human resource Machine Learning for Smart Recruitment
decisions. With a 93.2 percent accuracy rate and Systems," 2023 IEEE 3rd International
low errors (MAE 1.4, RMSE 2.0, and R 2 0.90), Conference on Artificial Intelligence and
the XGBoost model showed better performance. Computer Applications (ICAICA), 2023, pp.
The outcomes demonstrate that XGBoost is a 1-6.
[5]. S. Strohmeier and F. Piazza, "Artificial
great option for HR analytics and decision-
Intelligence Techniques in Human Resource
making tools since it identifies key performance Management—A Conceptual Exploration,"
variables and produces precise predictions. in Advanced Series in Management, vol. 15,
Compared to typical approaches and basic 2015, pp. 149-172.
regressions, machine learning, such as XGBoost [6]. M. Chen, H. Wang, and H. Ma, "Human
and Neural Networks, enables firms to measure Resource Management Under the Impact of
employee performance more objectively. The Artificial Intelligence," in Lecture Notes in
Computer Science, vol. 12066, 2020, pp.
HR teams can avoid making mistakes when
267-278.
making decisions by using the models, and they [7]. W. Park, "Artificial Intelligence and Human
even obtain reliable information that helps them Resource Management: New Perspectives
make decisions on labour management. AI and Challenges," 2018 International
models that use a combination of techniques to Conference on Computational Science and
acquire higher accuracies than XGBoost should Computational Intelligence (CSCI), 2018,
pp. 1361-1362.
be the focus of future research. Operators are in a
[8]. P. Domingos, "A Few Useful Things to
position to give better staff management when Know About Machine Learning,"
they assess employees based on their consistent Communications of the ACM, vol. 55, no.
performance, use machine learning to address 10, pp. 78-87, 2012.
poor performance swiftly, and offer one-on-one
[9]. B. Eubanks, "Artificial Intelligence For HR:
Use AI to Support and Develop a Successful
Workforce," Kogan Page, 2019.
[10]. I. Pradhan, "Impact of Artificial
Intelligence in Talent Acquisition Lifecycle
of Organization," International Research
Journal of Engineering and Technology
(IRJET), vol. 6, no. 8, pp. 1-5, 2019.
[11]. P. Matsa and K. Gullamajji, "To Study
Impact of Artificial Intelligence on Human
Resource Management," International
Research Journal of Engineering and
Technology (IRJET), vol. 6, no. 8, pp. 1-5,
2019.
[12]. M. Barabel, "L’IA au Service des RH
pour une Expérience Collaborateur
Augmentée," Revue de Gestion des
Ressources Humaines, vol. 120, pp. 129-
130, 2020.
