International Journal of Computer Science & Information Technology (IJCSIT) Vol 16, No 5, October 2024

A NEURAL NETWORK APPROACH TO

UNDERSTANDING EMPLOYEE RETENTION
DYNAMICS: INSIGHTS FROM FEATURE
IMPORTANCE ANALYSIS
Han-Chen Huang

Department of Tourism and MICE, Chung Hua University, Hsinchu, Taiwan.

ABSTRACT
In today’s business environment, employee retention has become a significant challenge, as employee
turnover can lead to decreased productivity, increased costs, and reduced morale. This study aims to
leverage neural network technology to predict employees' retention intentions and conduct in-depth
analyses based on employee data. A classification model was established considering various factors,
including gender, marital status, number of children, education level, years of service, weekly working
hours, career development opportunities, salary, and bonuses, to identify the potential risk of employee
turnover. The results indicated that the model achieved an accurate classification rate of 95.12% on the
test set, demonstrating high effectiveness in identifying employee retention intentions. Feature importance
analysis revealed that education level (29.63%), bonuses (27.50%), and the number of children (21.81%)
were the primary factors influencing retention decisions. Additionally, working hours, marital status, and
career development opportunities also impacted employees’ retention intentions. This research not only
provides insights into employee mobility but also offers data support for enterprises to develop effective
retention strategies, suggesting that businesses prioritize enhancing education and training opportunities,
designing competitive bonus systems, and providing flexible work arrangements and benefits tailored to
employees with children.

KEYWORDS
Neural Network Model, Employee Retention, Feature Importance Analysis

1. INTRODUCTION
Employee retention has emerged as a critical issue for modern organizations. High turnover rates
disrupt operations, escalate costs, and undermine team cohesion, ultimately affecting overall
productivity and employee morale [1-3]. As organizations strive to maintain a stable workforce,
human resource departments must proactively identify employees at risk of leaving. Timely
intervention is essential to implement effective retention strategies that can mitigate the negative
impacts of turnover [2-4]. Predicting employee behavior is inherently complex, influenced by
various factors including individual circumstances, organizational culture, and external economic
conditions. To address these challenges, organizations are increasingly leveraging advanced
technologies such as machine learning and artificial intelligence to enhance the accuracy of
retention predictions [5,6]. These technologies provide powerful tools to analyze vast datasets,
uncovering insights that traditional methods might overlook.

DOI: 10.5121/ijcsit.2024.16502 23
 International Journal of Computer Science & Information Technology (IJCSIT) Vol 16, No 5, October 2024

Neural networks, in particular, are gaining attention for their capacity to process large amounts of
data and identify intricate patterns within it [7,8]. By analyzing historical employee data, neural
networks can effectively model and predict employees' intentions regarding retention. This ability
to discern subtle correlations makes neural networks an appealing choice for tackling the
multifaceted issue of employee turnover [9-12].

The current study aims to develop a neural network model that accurately classifies employees'
intentions to stay or leave, utilizing actual employee data for model training and testing. This
approach will provide businesses with actionable insights to enhance employee retention rates,
ultimately fostering greater organizational stability and competitiveness.

2. LITERATURE REVIEW
The issue of employee retention has been widely studied in the fields of organizational behavior
and human resource management. According to Das (2013), effective employee retention
strategies are crucial for organizations seeking to maintain a competitive advantage in today’s
dynamic business environment. The author reviews a wealth of literature that emphasizes how
factors such as job satisfaction, work-life balance, and organizational culture significantly
influence employees' decisions to stay or leave [1]. This foundational work underscores the
necessity of addressing employee issues comprehensively to foster loyalty and reduce turnover.
Nair (2024) highlights the evolving nature of employee retention, suggesting that organizations
should reconsider their notice period policies as a strategic tool for retention. This paper
emphasizes that when employees feel that their departure would disrupt team dynamics, they may
be more likely to reconsider their decision to leave [2]. Similarly, Bajaj (2022) examines the
factors influencing employee retention among IT professionals in India, identifying key elements
such as career advancement opportunities, professional development, and job security that
significantly affect retention rates [3].

Shrestha and Prajapati (2023) provide insights into the relationship between human resource
management practices and employee retention. Their study shows that positive HR practices,
such as effective communication, recognition programs, and employee engagement initiatives,
significantly enhance employee loyalty and retention [4]. Moreover, Malik et al. (2020) explore
the role of perceived supervisor support in the relationship between HR practices and employee
retention, concluding that a supportive supervisory environment fosters greater employee
commitment and satisfaction [5].

The integration of advanced technologies, particularly AI-driven predictive analytics, is

becoming an important strategy for improving employee retention. Basnet (2024) discusses how
artificial intelligence can analyze vast amounts of employee data to identify potential turnover
risks, enabling organizations to take proactive measures to retain key talent [6]. Huang (2014)
and Liu and Lai (2021) emphasize the importance of artificial intelligence in demand forecasting
and advanced machine learning algorithms, such as convolutional neural networks, in enhancing
predictive capabilities across various domains, including employee retention [7,8].

Additionally, existing literature highlights the critical role of employee satisfaction in retention.
Al Kurdi et al. (2020) establish a theoretical and empirical foundation for understanding how
employee satisfaction directly correlates with customer satisfaction and organizational
performance, reinforcing the notion that satisfied employees are less likely to leave [9]. Insights
from Murtiningsih (2020) further support this idea, indicating that compensation, training, and
organizational culture are crucial in shaping job satisfaction and, consequently, employee
retention [10]. Ineson et al. (2013) and Jung and Yoon (2013) also contribute to this discussion,

24
 International Journal of Computer Science & Information Technology (IJCSIT) Vol 16, No 5, October 2024

emphasizing the significance of employee loyalty, indicating that loyalty directly impacts
customer satisfaction and repeat business [11,12].

The advancements in neural network models offer promising avenues for enhancing prediction
accuracy in employee retention. Studies by Dai (2024) and Jiang et al. (2024) explore the
implications of neural networks in forecasting employee behavior, suggesting that these models
can capture complex patterns that traditional methods may overlook [13,14]. Goodfellow et al.
(2016) and Hagan et al. (2014) provide foundational insights into the design and functionality of
neural networks, emphasizing their capabilities in handling intricate datasets [15,16]. These
advancements in deep learning technologies signify a shift towards more sophisticated, data-
driven approaches in understanding and improving employee retention strategies.

In conclusion, the literature presents a comprehensive view of employee retention, identifying

crucial factors and advanced methodologies that can enhance retention strategies. As
organizations continue to navigate the complexities of employee behavior, integrating advanced
predictive analytics and fostering a supportive work environment will be key to improving
employee retention.

3. RESEARCH METHODOLOGY
This study aims to apply neural networks to analyze employee data to predict retention intentions
and identify potential turnover risks. The research methodology includes dataset selection, data
preprocessing, model development, training and testing processes, parameter tuning, and
evaluation and analysis of results.

3.1. Dataset Description

The dataset comprises 198 employee records, with each record including 9 input features and 1
target variable. The input features include gender, marital status, number of children, education
level, years of service, working hours, career development opportunities, salary, and bonuses.
The target variable is the employee's retention intention (Stay or Leave) (Table 1). These features
cover multiple dimensions influencing employee retention or turnover, and the model aims to
predict employees’ intentions based on these features:

Table 1. Input and Output Features

Feature Description
Gender Employee's gender
Marital Status Employee's marital status
Number of Children Number of children of the employee
Education Level Employee's highest education level
Years of Service Number of years the employee has worked
Input
Working Hours Average weekly working hours of the employee
Career Development Number of career development opportunities
Opportunities within the company
Salary Average monthly salary of the employee
Average monthly bonuses of the employee in the
Bonuses
previous year
indicates the most probable employee intention:
Output Employee Intention
"Stay" or "Leave"

25
 International Journal of Computer Science & Information Technology (IJCSIT) Vol 16, No 5, October 2024

3.2. Model Training and Testing

The dataset was split into three subsets: a training set (58.58%), a validation set (20.71%), and a
test set (20.71%). This allocation ensures that the model learns on most of the data, is fine-tuned
using the validation set, and is finally evaluated on unseen test data. A backpropagation neural
network was employed to predict whether an employee would stay or leave. The network
structure includes an input layer (with 9 neurons corresponding to the 9 input features), 1 or 2
hidden layers (to capture complex relationships between features), and an output layer, used to
predict retention intentions (Stay or Leave).

The training process includes:

 Activation Functions: The hidden layers use an activation function, which is commonly
used for its ability to capture non-linear relationships. The output layer uses an
activation function, ideal for binary classification tasks, converting results to
probabilities between 0 and 1.
 Loss Function: The binary cross-entropy loss function was used, which is a standard
choice for binary classification problems.

4. MODEL DEVELOPMENT AND ANALYSIS

This study uses a neural network model to analyze employee retention intentions through precise
data modeling, aiming to reveal the key factors influencing employee retention. The model's
development process includes architecture design, parameter optimization, and performance
evaluation, providing a solid foundation for subsequent predictions and analyses.

4.1. Neural Network Architecture

This study employs a neural network-based model with the following specific settings [16-21]:

 Input activation function: Logistic

 Output error function: Cross-entropy
 Output activation function: Logistic
 Classification model: Confidence limits
 Acceptance threshold: 0.5
 Rejection threshold: 0.5

The network architecture consists of three layers:

 Input Layer: Comprising 17 input neurons corresponding to the 9 input features in the
dataset.
 Hidden Layer: Designed with one to two layers, with the number of neurons in each
layer optimized to improve model performance.
 Output Layer: Contains one output neuron responsible for binary classification
(predicting employee retention or turnover).

The optimal structure of the model was identified through exhaustive search methods, with the
hidden layer range set to 1-2 layers and evaluated based on test error. After iterations, the best
five network architectures were selected, as shown in (Table 2).

26
 International Journal of Computer Science & Information Technology (IJCSIT) Vol 16, No 5, October 2024
Table 2. Best Five Network Architectures

Architecture Akaike’s Criteria Fitness Test Error

17-5-1 -0.000958 41 0.97561
17-7-1 -0.001034 41 0.97561
17-16-1 -0.001599 41 0.97561
17-20-1 -0.002113 41 0.97561
17-21-1 -0.002297 41 0.97561

Akaike’s Criterion is a statistical criterion used for model selection. Proposed by Japanese
statistician Hirotugu Akaike in 1974 [22], the primary purpose is to select the model that best
explains the data among a set of models while considering model complexity. Among the
architectures, the 17-5-1 structure performed the best (with 17 input neurons, 1 hidden layer with
5 neurons, and 1 output neuron), with a fitness of 41, test error of 0.97561, and the lowest
Akaike’s Criteria value, thus selected as the optimal structure.

4.2. Model Training and Validation

The neural network model was trained using batch backpropagation, with both the hidden and
output layers employing logistic activation functions. The dataset was divided into training
(58.58%), validation (20.71%), and test sets (20.71%). Prediction errors were minimized by
adjusting model parameters, and classification accuracy was used to evaluate the model's
performance. The correct classification rate during the training phase reached 99.17%, with a
classification accuracy of 95% during the validation phase. Ultimately, the model achieved a
correct classification rate of 95.12% on the test set, demonstrating robust classification
capabilities. The following are the classification results for the test set (Table 3):

Table 3. Classification Results

Employee Predict Retention Intention

Retention Intention High Low Correct Predictions
Actual Retention High 30 1 30
Intention Low 1 9 9
Overall Correct Classification Rate: 95.12%

From the table, it can be observed that the model can accurately differentiate the majority of
employees' retention intentions. Among a total of 41 samples, only 2 samples were misclassified
(1 high-intention employee was classified as low intention, and 1 low-intention employee was
classified as high intention), resulting in an overall correct classification rate of 95.12%(=39/41).
These results indicate that the model can efficiently and accurately identify employees who may
leave. Identifying employees at risk of leaving is crucial in organizational management, allowing
businesses to take appropriate intervention measures in advance, such as providing better
development opportunities or adjusting compensation and benefits.

Further analysis of the impact of the nine input features on the model's output was conducted
using feature importance to measure each feature's contribution to the prediction results. The
importance percentages of each feature are presented in Table 4:

27
 International Journal of Computer Science & Information Technology (IJCSIT) Vol 16, No 5, October 2024
Table 4. Importance Percentages of Each Feature

Feature Importance (%)

Education Level 29.63
Bonus 27.50
Number of Children 21.81
Working Hours 7.97
Marital Status 5.63
Career Development Opportunities 4.66
Salary 2.15
Gender 0.03
Years of Service 0.63

From the table, it is evident that “Education Level” (29.63%), “Bonus” (27.50%), and “Number
of Children” (21.81%) have the most significant impact on employee retention decisions. This
suggests that enhancing employee education and creating competitive bonus systems could
improve retention rates. Specifically, increasing education levels not only aids in employees'
career advancement but also fosters greater loyalty to the organization. Companies should
consider providing ongoing training and development opportunities to support professional
growth.

Additionally, factors such as “Working Hours” (7.97%), “Marital Status” (5.63%), and
“Career Development Opportunities” (4.66%) also influence retention intentions. Although
these factors have a relatively smaller impact, they should not be overlooked. Reasonable
working hours and a good work-life balance can enhance employee satisfaction, thereby
influencing their retention decisions. Moreover, marital status may affect employees' expectations
and needs regarding work, which should be considered in the formulation of human resource
strategies.

Ultimately, these findings provide organizations with concrete suggestions for improving
employee retention, emphasizing that effective employee management should focus on creating a
supportive and rewarding work environment. Specifically, organizations should pay attention to
employees' individual needs and career development and implement corresponding incentive
measures to enhance retention rates and overall job satisfaction.

5. CONCLUSION
This study demonstrates the potential of neural networks in predicting employee retention
intentions through a comprehensive analysis of employee data. By effectively classifying
employees based on various features, the model provides valuable insights into retention risks.
The achieved overall correct classification rate of 95.12% reflects the model's strong performance
in accurately identifying employees' intentions to stay or leave.

Feature importance analysis reveals that education level, bonuses, and the number of children
significantly influence employee retention decisions. Organizations can utilize these findings to
develop targeted retention strategies, such as enhancing training opportunities, improving
compensation packages, and tailoring benefits to employees with families.

28
 International Journal of Computer Science & Information Technology (IJCSIT) Vol 16, No 5, October 2024

In conclusion, neural networks are a powerful tool for predicting employee retention, offering
businesses actionable insights that can lead to improved employee satisfaction, reduced turnover,
and enhanced organizational performance. Future research could explore additional variables and
advanced model architectures to further improve prediction accuracy and deepen understanding
of employee retention dynamics.

REFERENCES

[1] Das, Bidisha. (2013). Employee Retention: A Review of Literature. IOSR Journal of Business and
Management. 14, 8-16. https://doi.org/10.9790/487X-1420816.
[2] Nair, Manju. (2024). Employee retention a top priority and what if it doesn’t work: Time to look at an
ideal notice period. Human Systems Management. 43(1), 1-15. https://doi.org/10.3233/HSM-230220.
[3] Bajaj, K.K. (2022). Investigation into Factors Influencing Employee Retention Among IT
Professionals: A Perspective from India. Journal of Survey in Fisheries Sciences. 8(3), 457-467.
https://doi.org/10.53555/sfs.v8i3.2441.
[4] Shrestha, Prakash and Prajapati, Dhan. (2023). Human Resource Management Practices and
Employee Retention. Journal of Balkumari College. 12, 1-9.
https://doi.org/10.3126/jbkc.v12i1.60415.
[5] Malik, E., Baig, S. A., and Manzoor, U. (2020). Effect of HR practices on employee retention: The
role of perceived supervisor support. Journal of Public Value and Administrative Insight, 3(1), 1-7.
[6] Basnet, Sunil. (2024). The Impact of AI-Driven Predictive Analytics on Employee Retention
Strategies. International Journal of Research and Review. 11, 50-65.
https://doi.org/10.52403/ijrr.20240906.
[7] Huang, Han-Chen. (2014). A study on artificial intelligence forecasting of resort demand. Journal of
Theoretical and Applied Information Technology, 70, 265-272.
[8] Liu, Kuang-Tai, and Lai, Yu-Ting. (2021, May 21). Defect Detection in Coffee Beans Using the
YOLO Module and Convolutional Neural Networks. In The 2021 Conference on Enterprise
Competitiveness & Management (pp. 410-416). Chung Hua University, Hsinchu, Taiwan.
[9] Al Kurdi, Barween and Alshurideh, Muhammad & Alnaser, Ahmad. (2020). The impact of employee
satisfaction on customer satisfaction: Theoretical and empirical underpinning. Management Science
Letters. 3561-3570. https://doi.org/10.5267/j.msl.2020.6.038.
[10] Murtiningsih, Retno. (2020). The Impact of Compensation, Training Development, and
Organizational Culture on Job Satisfaction and employee Retention. Indonesian Management and
Accounting Research. 19(1), 33-50. https://doi.org/10.25105/imar.v19i1.6969.
[11] Ineson, E., Benke, E., & László, J. (2013). Employee loyalty in Hungarian hotels. International
Journal of Hospitality Management, 34(1), 31-39. https://doi.org/10.1016/j.ijhm.2012.04.001.
[12] Jung, Hyo and Yoon, Hye Hyun. (2013). Do employees’ satisfied customers respond with a
satisfactory relationship? The effects of employees’ satisfaction on customers’ satisfaction and loyalty
in a family restaurant. International Journal of Hospitality Management. 34, 1-8.
https://doi.org/10.1016/j.ijhm.2013.02.003.
[13] Dai, Ying. (2024). Predicting the Reasons of Employee Turnover Based on BP Neural Network
Model. Highlights in Science, Engineering and Technology. 100, 119-127.
https://doi.org/10.54097/x2vjgb22.
[14] Jiang, Chunheng and Huang, Zhenhan & Pedapati, Tejaswini & Chen, Pin-Yu & Sun, Yizhou & Gao,
Jianxi. (2024). Network properties determine neural network performance. Nature Communications.
15, Article 48069. https://doi.org/10.1038/s41467-024-48069-8.
[15] Goodfellow, I., Bengio, Y., & Courville, A. (2016). Deep learning. MIT Press.
[16] Hagan, M. T., Demuth, H. B., Beale, M. H., & De Jesús, O. (2014). Neural network design (2nd ed.).
Martin Hagan.
[17] Ma, Heng, and Lin, Shih-Hung. (2021, May 21). A Study on Algorithm Strategies and Long-Term
Holdings: A Case Study of E.SUN Financial Holding Company. In The 2021 Conference on
Enterprise Competitiveness & Management (pp. 213-219). Chung Hua University, Hsinchu, Taiwan.
[18] Liu, Kuang-Tai, and Tseng, Tzu-Ling. (2021, May 21). Pineapple Leaf Image Recognition Using
Convolutional Neural Networks. In The 2021 Conference on Enterprise Competitiveness &
Management (pp. 231-236). Chung Hua University, Hsinchu, Taiwan.

29
 International Journal of Computer Science & Information Technology (IJCSIT) Vol 16, No 5, October 2024
[19] Liu, Kuang-Tai, and Hsu, Chia-Hsuan. (2021, May 21). Recognition of Healthy Banana Leaves Using
Convolutional Neural Networks. In The 2021 Conference on Enterprise Competitiveness &
Management (pp. 535-540). Chung Hua University, Hsinchu, Taiwan.
[20] Che, Hui-Chung, Lai, Yi-Hsuan, and Wang, Szu-Yi. (2010). Assessment of Patent Legal Value by
Regression and Back-Propagation Neural Network. International Journal of Systematic Innovation,
1(1), 31-47.
[21] Lai, Yi-Hsuan, and Che, Hui-Chung. (2009). Modeling Patent Legal Value by Extension Neural
Network. Expert Systems with Applications, 36(7), 10520-10528.
https://doi.org/10.1016/j.eswa.2009.02.072.
[22] Akaike, H. (1974). A new look at the statistical model identification. IEEE Transactions on
Automatic Control, 19(6), 716-723.

30