Thanks to visit codestin.com
Credit goes to www.scribd.com

0% found this document useful (0 votes)
5 views9 pages

Article

This study presents a Bi-LSTM deep learning approach for predicting employee churn, achieving an accuracy of 97.5%. It compares the effectiveness of Bi-LSTM with other machine learning models such as Logistic Regression and Support Vector Machine, concluding that Bi-LSTM outperforms them. The research emphasizes the importance of timely employee churn prediction to mitigate organizational losses associated with turnover.

Uploaded by

meriem.eloualfi
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
5 views9 pages

Article

This study presents a Bi-LSTM deep learning approach for predicting employee churn, achieving an accuracy of 97.5%. It compares the effectiveness of Bi-LSTM with other machine learning models such as Logistic Regression and Support Vector Machine, concluding that Bi-LSTM outperforms them. The research emphasizes the importance of timely employee churn prediction to mitigate organizational losses associated with turnover.

Uploaded by

meriem.eloualfi
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 9

Open Access ORIGINAL ARTICLE

Full Length Article

Bi-LSTM Deep Learning Approach for Employee Churn


Prediction
Madiha Qadir1, Iram Noreen2, Asghar Ali Shah3
1Department of Computer Sciences, Bahria University Islamabad, Lahore Campus, 54000, Pakistan
2Department of Computer Sciences, Bahria University Islamabad, Lahore Campus, 54000, Pakistan
3Department of Computer Sciences, Bahria University Islamabad, Lahore Campus, 54000, Pakistan

ABSTRACT
Employee churn prediction also known as ‘attrition’ or ‘turnover’ is referred to as the identification of employees planning
to quit the organization in the future. Organizations invest time, effort, and money in employees’ training. Therefore, an
experienced employee is an asset to the organization. If organizations could predict employee churn using machine
learning techniques and can take timely measures, then they can prevent long-term loss. Several machine learning
models have been used for churn prediction of employees, such as Logistic Regression, Support Vector Machine, and
MLP (Multi-Layer Perceptron). This study aims to find the optimal algorithm of classification for the prediction of the
churn employee rate. A deep learning approach based on B-LSTM (Bi-Directional Long Short-Term Memory) is being
proposed and tested. The accuracy of B-LSTM is 97.5% during the consistency test. A comparative analysis with other
machine-learning techniques is also performed and it is concluded that B-LSTM has proved more effective than other
machine learning techniques investigated in this study.
Keywords: Churn Prediction, MLP (Multi-Layer Perceptron), Bidirectional LSTM (B-LSTM), Gradient Boosting, Naïve
Bayes, Organization
Author`s Contribution Address of Correspondence Article info.
1 Iram Noreen
Data analysis, interpretation and Received: January 07, 2021
manuscript writing, Active participation in Email: [email protected] Accepted: May 11, 2021
data collection, 2Conception, synthesis, Published: June 30, 2021
planning of research, 3. Interpretation and
discussion
Cite this article: Qadir M, Noreen I, Shah AA. Bi-LSTM Deep Learning Approach for Funding Source: Nil
Employee Churn Prediction. inf. commun. technol. robot. appl.2021; 12(1):1-10. Conflict of Interest: Nil

INTRODUCTION
Churn prediction is referred to as the identification employees planning to quit the organization in the future
of persons planning to quit an organization process in the [4]. Human asset is the most critical and vulnerable asset
future [1-3]. Churn prediction has two types, customer of organizations. In a competitive and rapidly changing
churn, and employee churn. Churn customer is a business era, employees tend to find better job
prediction of when and which customer stops buying opportunities for growth and they leave their organization
products. This study focuses on employee churn easily for a better perspective. Churn is very difficult to
prediction. Employee churn prediction is also known as control and it is a nearly unavoidable process in any field.
‘attrition’ or ‘turnover’ is referred to as the identification of Therefore, organizations are

Journal of Information Communication Technologies and Robotic Applications


http://www.jictra.com.pk/index.php/jictra, pISSN: 2523-5729, eISSN: 2523-5739
JICT R A 2 021 1
focusing nowadays on prior identification of employees The main contributions of this paper are summarized as
who might have a high tendency to leave [1, 5, 6]. follows:
Organizations taking timely measures to reduce churn  A deep learning approach based on RNN/LSTM is
problems are more effective in executing long-term proposed and implemented by using bi-directional
business plans. Employee churn prediction could be properties to predict the optimal accuracy of churn
helpful for an organization to avoid a high churn rate of employees.
valuable employees [7]. It is affected by several factors  Multiple state of art machine learning approaches
such as growth potential, pays, promotions, working such as MLP and Naïve Bayes is implemented for
environment, working hours, and job satisfaction [8]. better implementation with parameter tuning.
Some employees leave involuntary for reasons like  Quantitative performance comparison among all
retirements, internal transfers, and performance issues [9- implemented approaches is presented.
12]. Hiring new employees to replacement of churn The rest of the paper is organized as follows: Section 2
employees costs a lot in terms of training and hiring describes the literature review. Section 3 explains the
process [13, 14]. The new employee requires more time research methodology. Section 4 presents the
to learn processes and gain experience. It affects the experimental results and discussions followed by the
performance of other employees as well. They get conclusion in Section 5.
demotivated by the unexpected resignation of a senior co-
BACKGROUND
worker. Similarly, working in an organization for a long
time without any increments also pushes the workers to In literature, several researchers have investigated the
resign after some duration. This phenomenon makes problem of employee churn. In a study, the XGBoost
organizations suffer in the long term as they invest time, classifier is used which proved to be a superior algorithm
human effort, and money to train their workers and with high accuracy and low runtime, and it predicted
convert them into organizational assets [15]. Employee employee turnover with 95% accuracy [18]. In another
churn is the main hassle in telecom, IT, and many other study, Yigit et al. [22] evaluated multiple classifiers and
sectors now a day [16]. Therefore, employers want to presented results as logistic regression with 87.1%
retain employees and want to prevent worker turnover. and SVM 89.7% accuracy rate. Naïve Bayes is also used
Several machine learning classification models have been in another study by Yedida et al. [17] with 85.6%
used for churn prediction such as gradient boosting, accuracy. Feature selection methods can be utilized to
Naïve Bayes, MLP (Multi-Layer Perceptron), logistic build many models with special subsets of train datasets
regression, SVM (Support Vector Machines), KNN (K- and decide these aspects that are and are no longer
Nearest Neighbor), Random Forest, Decision Trees [6, relevant to building a dependable and accurate model
15-24]. However, the accuracy of recent churn prediction [25]. Alamsyah et al. [17] conducted a survey that
techniques is not optimal and there is room for includes 38 questions like overall satisfaction,
improvement. Deep learning is an emerging subset of loyalty, motivation, gender, age, and salary is
machine learning approaches that need to be extensively highlighted as key distinguished parameters helpful for
explored yet to resolve the employee churn problem. This the training of machine learning models to predict
research has investigated the behavior of the above- employee churn. They have used Naïve Bayes-based
mentioned classification models using public research prediction model with 10-fold cross-validation for
dataset of employees. Further, the potential of deep evaluation.
learning is explored for employee churn prediction and a In another study, C4.5 classifiers are used with 90%
novel deep learning-based approach is proposed for training data and 10% testing data. They evaluated the
churn prediction with enhanced performance and accuracy of C4.5 classifiers as 77 % [26]. In another
accuracy. Moreover, a comparative analysis between the study, the classification technique used is based on 10-
above-mentioned models and the proposed model is also fold cross-validation training and a test dataset. They
presented. used data mining tools such

Journal of Information Communication Technologies and Robotic Applications


http://www.jictra.com.pk/index.php/jictra, pISSN: 2523-5729, eISSN: 2523-5739
JICT R A 2 021 2
as WEKA and ROSETTA toolkit in their experiment [26, proposed an approach to highlight the factors of customer
27]. Their experiment resulted in 95% accuracy to train churn using the attribute-selection classifier algorithm.
C4.5 classifiers. Sisodia et al. trained the KNN model on Two datasets have been used. They used Random
the HR dataset with 96% accuracy [28]. Naïve Bayes forests with 95% accuracy [38]. In another study bagging,
classifier reported 72.7% accuracy and SVM reported MLP is used on two benchmark datasets obtained from
51.2% accuracy on customer churn data of Twitter [29]. In GitHub for evaluation of the proposed model with 94%
another study, traditional machine learning predictive accuracy [39]. In another study, the dataset is collected
models did not perform well in predicting the churn of from one of China’s largest online professional social
workers however, they pointed out the factors contributing network websites. The dynamic Bipartite Graph
to the worker’s churn or worker’s motivations for staying Embedding (DBGE) method is used, which learns low-
longer in the organization [30]. In another study, a method dimensional vector representations. They acquired 89%
to approach worker retention has been proposed using accuracy by applying DBGE random forest [40]. Jain et al.
general machine learning strategies such as SVM and it [33] used datasets of banking, telecom, and IT to predict
gives a ROC of 80%. They observed and reported that for employee churn. They acquired 86.3% accuracy in the
the information at hand, the maximum efficient retention banking sector using random forest, 82.9% accuracy in
exercise became to broaden skills mobility throughout the telecom sector using XGBoost, and 90% accuracy in
positions [31]. Further, other strategies such as decision the IT sector using logistic regression. Lalwani et al. [41]
tree, SVM, and neural network are also investigated using applied ANN for employee churn and acquired 86.5%
an open-source software program called WEKA [32] and accuracy. Castellini et al collected employee data from an
their experiment reported performance of decision tree at American Telecom Company named ‘Orange’ and
77.9%, neural networks at 83.7%, and SVM at 83.7% acquired 85.2% accuracy using logistic regression.
accuracy. A method to approach worker retention has Khodadadi et al. [42] acquired 95% accuracy using
been proposed using general machine learning strategies XGBOOST on structured and semi-structured data of the
such as logistic regression accuracy is 90.1% in IT sector, company. Jain et al. [43] trained logistic regression using
telecom sector has 82.9% and in banking sector 86.3% a telecom dataset with 80% accuracy. In another study, a
accuracy [33]. In another study, SVM accuracy is 70.9%, large amount of professional social data from one of the
Naïve Bayes has 77.4% and random forest has acquired largest workplace social platforms in China used Cox-
77.4% accuracy [34]. Random Forest with 85.6% [44]. Hu et al. [45] acquired
Yang et al. proposed an approach for churn time 81% accuracy training Adaboost classifier on customer
prediction of players using linear regression. They used dataset.
the datasets of six free online games: Thirty-six
RESEARCH METHODOLOGY
Stratagems (TS), Thirty-six Stratagems Mobile (TSM),
Game of Thrones Winter is Coming (GOT), Woman land Dataset Acquisition and Pre-processing
in Journey to the West (WJW), League of Angels II (LOA An employee churn dataset named HR_comma_sep.csv
II), and Era of Angels (EOA). Their proposed approach dataset [8] from Kaggle is used. It comprises fifteen
acquired 78.9% accuracy [35]. Khalid et al. proposed an thousand (15,000) employees’ samples and provides ten
approach to attract future customers and to investigate (10) features/attributes. It includes features like
churning rate of current customers. They employed their satisfaction, evaluation, project count, average monthly
investigation on employee data of the Telecom industry hours, years at the company, work accident, promotion,
using Decision Tree to prevent the loss of potential department, turnover, and salary as shown in Table 1.
customers while retaining the happiness level of the Department and salary were in strings then convert into
current customers with 94% accuracy [36]. numbers through a label encoder. Department has 10
Li et al. used Gradient Boosting Decision Tree (GBDT) categories and salary has 3 categories including low,
and applied hyper-parameter tuning to acquire 94% medium and high.
accuracy for churn prediction [37]. Kassem et al. used

Journal of Information Communication Technologies and Robotic Applications


http://www.jictra.com.pk/index.php/jictra, pISSN: 2523-5729, eISSN: 2523-5739
JICT R A 2 021 3
Label encoder is used for feature scaling. Normalization is records through its memory and it’s efficient due to its
applied using the Min-Max Scaler method. We have used internal memory. RNNs are used to understand records in
an automated convenient feature selection technique sequential traits. They use patterns to predict the
Principal Component Analysis (PCA), which is a subsequent probable scenario. RNNs are widely used in
dimensionality reduction method to predict the accurate developing deep learning models. Its bidirectional model
result of employee churn. The assumption in deep has applications in speech recognition and prediction.
learning relies on the principle that data was generated by Recurrent neural networks are good to remember
a composition of factors that can be represented in some important things about the input received, which enables
hierarchical form. them to predict what is coming next. RNN produces
Proposed Approach predictive consequences sequentially that other
A deep learning model based on B-LSTM and LSTM is algorithms cannot perform [24].
proposed. LSTM [46] are modified form of the RNN
model. A recurrent neural network (RNN) is an effective
and well-known type of neural network. It can remember
Table 1. Details of HR Dataset
S. No. Attributes Data Type Value Range
1 Satisfaction Numeric 0.38-0.9
2 Evaluation Numeric 0.53-1
3 Project Count Numeric 2-7
4 Average Monthly hours Numeric 157-306
5 Years at Company Numeric 3-6
6 Work Accident Numeric 0-1
7 Turnover Numeric 1-0
8 Promotion Numeric 0-1
9 Department Categorical Sales(Different other department)
10 Salary Categorical Low/High/Medium

RNN has three gates; input gate, forget gate, and output process. It has at least one feedback loop. RNN can
gate. Forget gate finds out values to be discarded from model the sequence of data so that each sample can be
the block. It is determined by the sigmoid function. assumed to be dependent on previous ones. RNN is even
The input and the memory of the block are used to figure used with convolutional layers to extend the effective pixel
out the output. It trains the model by back-propagation of neighborhood. However, it cannot process long
weights and bias. It can process data from initial input to sequences if used with the ReLU activation function.
final output. It has a forward as well as a backward

Figure 1. Bidirectional RNN Framework [47]

Journal of Information Communication Technologies and Robotic Applications


http://www.jictra.com.pk/index.php/jictra, pISSN: 2523-5729, eISSN: 2523-5739
JICT R A 2 021 4
A Bidirectional RNN known as LSTM is a sequence Detailed Architecture
processing model comprising of two RNNs: one takes the The architecture of the proposed model is defined as a
input in a forward direction, and another one takes it in a total number of neurons in the input layer being 9, one
backward direction. It duplicates the first recurrent layer hidden layer includes 21 neurons and one neuron in the
in the network to develop two layers side-by-side. Then output layer. Adam optimization algorithm is used to find
first input sequence is provided as input to the first layer the weights and the accuracy metric is calculated and
and a reversed copy of the input sequence is provided to reported each epoch. Min-max scaler is used for
the second layer. In contrast to standard feed-forward normalization. ReLU is used as a classification function in
neural networks, LSTM has feedback connections. This the hidden layer. Sigmoid is used as an activation function
enables it not only to process single data points, but also in the hidden layer. The B-LSTM is trained for 1000
entire sequences of data. Bidirectional LSTM works in epochs. BI-LSTM block diagram with input layers, hidden
both directions forward as well as backward. It has input, layers, and an output layer which is shown in Figure 2.
hidden and output layers which are shown in Figure 1
[47].

Figure 2. BI-LSTM Structure

The bottom LSTM nets are used for the forward feature.
The top LSTM nets are used for backward. The two
networks connect at a common activation layer to
produce outputs. Neurons in a forward state of BLSTM
behave like a unidirectional LSTM structure. Thus, the
neurons in both networks are not directly connected as
shown in Figure 2. A backward pass is performed for
output neurons and the forward states. Finally, all weights
are revised and updated. Hence, BLSTM structures
provide better results than other network structures.
Internal architectural detail is shown in Figure 3.
Figure 3. Architecture Diagram

Implementation and Experimental Setup

Journal of Information Communication Technologies and Robotic Applications


http://www.jictra.com.pk/index.php/jictra, pISSN: 2523-5729, eISSN: 2523-5739
JICT R A 2 021 5
Google COLAB Python3 is used for implementation. different algorithms on our dataset by using COLAB.
TensorFlow's deep learning framework is used for These accuracies are better than the previous accuracy
implementation. RNN (Bidirectional LSTM) has been used found by different researchers.
to implement on the same dataset. The test train split ratio
is 70%-30%. Further, during training 10 Fold Cross-
validation is applied to manage to overfit.

RESULTS AND DISCUSSION


After the training a testing phases, results are analyzed by
following metrics to evaluate the performance of the
proposed model and previous methods: -
Accuracy: It is the number of correct predictions divided
by the total number of predictions.

..… .(1)
Precision: It is the ratio of correctly predicted positive
Figure 4. Confusion Matrix of Bidirectional LSTM
observations to the total predicted positive observations.
The confusion matrix of Bidirectional LSTM true positive is
………. (2) 911 which is the percentage of actual positives which are
F1-Score: The F1 Score is the weighted average of correctly identified, the true negative is 3417 which is the
precision and recall. Therefore, this score takes both false percentage of actual negatives which are correctly
positives and false negatives. identified, false positive is 45 which incorrectly predicts
the positive class and false negative is 127 which is
…..…. (3) incorrect which is shown in Figure 4.
Matthew’s Correlation Coefficient: The Matthews
correlation coefficient is used in machine learning as a
True / Actual
measure of the quality of binary classification.
No Yes

No TN = 98.7% FP = 1.29%
Predicted

……... (4)
10 Folds are used for cross-validation by partitioning the
Yes FN = 12.2% TP = 87.7%
training set to train the model, and a test set to evaluate it.
The training set is the one on which we train and fit our
Figure 5. Confusion Matrix of Bidirectional LSTM in
model basically to fit the parameters whereas test data is Percentage
used only to assess the performance of the
model. Training data's output is available to the model
whereas testing data is the unseen data for which Confusion matrix of Bidirectional LSTM true positive is
predictions have to be made. The training part of the 87.7% which is the percentage of actual positives which
dataset comprises 70% and the remaining 30% of the are correctly identified, the true negative is 98.7% which is
dataset is preserved for a testing phase in independent the percentage of actual negatives which are correctly
testing. By using similar data for training and testing, identified, false positive is 1.29% which incorrectly
you can minimize the effects of data discrepancies and predicts the positive class and false negative is 12.2%
better understand the characteristics of the model. After a which is incorrect which is shown in Figure 5.
model has been processed by using the training set,
you test the model by making predictions against
the test set. We find the accuracy and other parameters of

Journal of Information Communication Technologies and Robotic Applications


http://www.jictra.com.pk/index.php/jictra, pISSN: 2523-5729, eISSN: 2523-5739
JICT R A 2 021 6
Figure 6. Comparison of Consistency Test Parameters

A consistency test is performed when the model is fully performance. The closer the curve comes to the 45-
trained on given training data. Inconsistency test accuracy degree diagonal of the ROC space, the less accurate the
of B-LSTM is 97.5%, specificity is 98%, precision is 96%, model is. The ROC curve of B-LSTM is shown in Figure 6.
sensitivity is 92%, Mathew’s correlation coefficient is 93%, Its ROC is 95% and it shows the true positive and false
and F1 score is 94% as shown in Figure 6. positive rate of B-LSTM for an independent test. It means
The Receiver Operating Characteristic (ROC) curve is a there is a 95% chance that the model will be able to
trade-off between sensitivity and specificity and tells about distinguish between positive class and negative class
the true positive rate and false-positive rate. If it is nearest which is shown in Figure 7.
to the diagonal, then it is not a good curve. Classifiers that
give curves closer to the top-left corner indicate better

Figure 7. B-LSTM Consistency Test ROC Curve


CONCLUSION
Journal of Information Communication Technologies and Robotic Applications
http://www.jictra.com.pk/index.php/jictra, pISSN: 2523-5729, eISSN: 2523-5739
JICT R A 2 021 7
11. S. K. a. R. Vijay, "Job Satisfaction – A Major Factor Behind
Attrition or Retention in Retail Industry," Imperial Journal of
The prediction of employee churn is useful for HR of any Interdisciplinary Research, vol. 2, 2016.
organization to take necessary action for the retention of 12. M. H. K. S. Khan, A. M. Mohmand, and S. Misbah, "Impact of HR
employees predicted to be at risk of leaving. Machine practices on employee turnover and job satisfaction: Evidence
from Pakistani universities," Rev. Econ. Dev. Stud, vol. 6, pp.
learning algorithms like Naïve Bayes, MLP, and a 607–624, 2020.
proposed model based on BILSTM are used to predict the 13. A. Frederiksen, "Job Satisfaction and Employee Turnover, A
Firm-Level Perspective," German Journal Human Resource
employee churn problem. We have implemented the Management, vol. 31, pp. 132–161, 2017.
above-mentioned algorithms and Bidirectional LSTM (B- 14. P.-L. D. I.-C. Huang, L.-F. Wu, J. Achyldurdyyeva, L.-C. Wu, and
LSTM). B-LSTM’s accuracy is 97.5%, sensitivity is 92% C. S. Lin, "Leader-member exchange, employee turnover
intention, and presenteeism: the mediating role of perceived
specificity is 98%, precision is 96%, F1 score is 94% and organizational support," Leadership & Organization Development
MCC is 93%. On a larger dataset, B-LSTM will outperform Journal, 2021.
15. J. W. X. Gao, and C. Zhang, "An Improved Random Forest
even better because deep learning-based approaches Algorithm for Predicting Employee Turnover," Mathematical
need huge data sets to reach their full potential. Problems in Engineering, vol. 2019, pp. 1–12, 2019.
16. S. K. a. L. Shyamala, "Prediction of Employee Turnover Using
Therefore, it is evident that B-LSTM will be more effective
Ensemble Learning," Advances in Intelligent Systems and
in if huge training is provided either using data Computing Ambient Communications and Computer Systems,
augmentation techniques or a larger dataset is available. pp. 319–327, 2019.
17. R. R. Yedida, R. Vahi, R. Jana, Abhilash, D. Kulkarni, "Employee
Further experiments are planned as future work to Attrition Prediction," 2018.
discover the full potential and performance optimality of 18. R. P. a. P. Ajit, "Prediction of Employee Turnover in
Organizations using Machine Learning Algorithms," International
the proposed B-LSTM-based model by applying text
Journal of Advanced Research in Artificial Intelligence, vol. 5,
augmentation techniques to the employee dataset. 2016.
Data Availability 19. M. K. H. Y. Zhao, F. Cheng, B. Fu, and X. Zhu, "Employee
Turnover Prediction with Machine Learning: A Reliable
The data set download link is already shared in Approach," Advances in Intelligent Systems and Computing
references. Model implementation code and detailed Intelligent Systems and Applications, pp. 737–758, 2018.
20. S. A. E. A. el Kassem, A. Mostafa, and F. Kamal, "Customer
experimental results will also be available soon on Github. churn prediction model and identifying features to increase
customer retention based on user-generated content," Int. J. Adv.
REFERENCES Comput. Sci. Appl., vol. 11, 2020.
21. V. C. N. a. K. Dimililer, "Customer churn prediction for business
1. P. L. E. Zdravevski, C. Apanowicz, and D. Ślȩzak, "From Big
intelligence using machine learning," presented at the 3rd
Data to business analytics: The case study of churn prediction," International Congress on Human-Computer Interaction,
Appl. Soft Comput., vol. 90, p. 106164, 2020. Optimization and Robotic Applications (HORA), 2021.
2. S. Younis, and Ali Ahsan, "Know Your Stars Before They Fall 22. I. O. Y. a. H. Shourabizadeh, "An Approach for Predicting
Apart: A Social Network Analysis of Telecom Industry to Foster Employee Churn by Using Data Mining," International Artificial
Employee Retention Using Data Mining Technique," IEEE Intelligence and Data Processing Symposium, pp. 1-4, 2017.
Access vol. 9, pp. 16467-16487, 2021. 23. W.-C. Y. S. Wu, T.-S. Ong, and S.-C. Chong, "Integrated churn
3. N. B. Yahia, Jihen Hell, and Ricardo Colomo-Palacios, "From Big prediction and customer segmentation framework for telco
Data to Deep Data to Support People Analytics for Employee business," IEEE Access, vol. 9, pp. 62118–62136, 2021.
Attrition Prediction," IEEE Access, vol. 9, pp. 60447-60458, 2021. 24. M. P. Z. J. M. D. Oliveira, P. A. Gloor, and T. Joshi, "Mirror,
4. V. V. S. a. G. K. Palshikar, "Employee Churn Prediction," Expert Mirror on the Wall, Who Is Leaving of Them All: Predictions for
System with Application, vol. 38, pp. 1999-2006, 2011. Employee Turnover with Gated Recurrent Neural Networks,"
5. A. J. A. K. Ahmad, and K. Aljoumaa, "Customer churn prediction Studies on Entrepreneurship, Structural Change and Industrial
in telecom using machine learning in big data platform," J. Big Dynamics Collaborative Innovation Networks, pp. 43–59, 2019.
Data, vol. 6, 2019. 25. F. Y. X. Lin, L. Zhou, P. Yin, H. Kong, W. Xing, X. Lu, L. Jia, Q.
6. S. M. a. D. Chitre, "A survey of employee and customer churn Wang, and G. Xu, "A Support Vector Machine-Recursive Feature
prediction methodologies," Adv. Math. Sci. J., vol. 9, pp. 3955– Elimination Feature Selection Method Based on Artificial Contrast
3962, 2020. Variables and Mutual Information," Journal of Chromatography B,
7. H. Ongori, "A Review of the Literature on Employee Turnover," vol. 910, pp. 149-155, 2012.
African Journal of Business Management, vol. 1, pp. 49-54, 2007 26. A. R. H. H. Jantan, Z. A. Othman, "Human Talent Prediction in
8. Kaggle, "hr-comma-sep," Kaggle, Ed., ed, 2019. HRM using C4.5 Classification Algorithm," International Journal
9. H. H. a. E. A. Garcia, "Learning from Imbalanced Data," IEEE on Computer Science and Engineering (IJCSE), vol. 2, pp. 2526-
Transactions on Knowledge and Data Engineering, vol. 21, pp. 2534, 2010.
1263-1284, 2009. 27. A. R. H. Jantan, and Z. Ali, "Data Mining Classification
10. D. M. Powers, "Evaluation: From Precision, Recall and F- Techniques for Human Talent Forecasting," knowledge-oriented
measure to ROC and Correlation," Journal of Machine, 2011. Applications in Data Mining, 2011.

Journal of Information Communication Technologies and Robotic Applications


http://www.jictra.com.pk/index.php/jictra, pISSN: 2523-5729, eISSN: 2523-5739
JICT R A 2 021 8
28. S. V. D. S. Sisodia, and A. Pujahari, "Evaluation of Machine 45. Y. Z. Z. Hu, and M. Khushi, "A survey of Forex and stock price
Learning Models for Employee Churn Prediction," presented at prediction using deep learning," Appl. Syst. Innov., vol. 4, p. 9,
the International Conference on Inventive Computing and 2021.
Informatics (ICICI), 2017. 46. Ö. Yildirim, "A Novel Wavelet Sequence Based on Deep
29. N. A. S. P. M. D. N. Arusada, and A. Alamsyah, "Training Data Bidirectional LSTM Network Model for ECG Signal
Optimization Strategy for Multiclass Text Classification," Classification," Computers in Biology and Medicine, vol. 96, pp.
presented at the 5th International Conference on Information and 189–202, 2018.
Communication Technology (ICICT), 2017. 47. N. J. A. Graves, and A.-R. Mohamed, "Hybrid speech recognition
30. A. J. S. Yadav, and D. Singh, "Early Prediction of Employee with Deep Bidirectional LSTM," IEEE Workshop on Automatic
Attrition using Data Mining Techniques," presented at the IEEE Speech Recognition and Understanding, pp. 273-278, 2013.
8th International Advance Computing Conference (IACC), 2018.
31. K. T. E. Ribes, B. Perthame, "Employee Turnover Prediction and
Retention Policies Design," 2017.
32. Y. H. E. Shaaban, A. Khedr, M. Nasr, "A Proposed Churn
Prediction Model," International Journal of Engineering Research
and Applications (IJERA), vol. 2, pp. 693-697, 2012.
33. G. Y. H. Jain, and R. Manoov, "Churn prediction and retention in
banking, telecom and IT sectors using machine learning
techniques," in Advances in Machine Learning and
Computational Intelligence, 2021, pp. 137–156.
34. N. B. a. A. Jadhav, "Churn Prediction of Employees using
Machine Learning Techniques," The. glas. , vol. 15, pp. 51–59,
2021.
35. T. H. W. Yang, J. Zeng, L. Chen, S. Mishra, and Y. Liu, "Utilizing
Players Playtime Records for Churn Prediction: Mining Playtime
Regularity," IEEE Transactions on Games, 2020.
36. A. M. A. L. F. Khalid, D. Q. Zeebaree, F. Y. H. Ahmed, and D. A.
Zebari, "Customer churn prediction in telecommunications
industry based on data mining," presented at the IEEE
Symposium on Industrial Electronics & Applications (ISIEA),
2021.
37. J. Li, Lu, H., Wang, C., Ma, W., Zhang, M., Zhao, X., "A difficulty-
aware framework for churn prediction and intervention in games,"
in Proceedings of the 27th ACM SIGKDD Conference on
Knowledge Discovery & Data Mining, 2021, pp. 943-952.
38. B. R. I. Ullah, A. K. Malik, M. Imran, S. U. Islam, and S. W. Kim,
"A Churn Prediction Model Using Random Forest: Analysis of
Machine Learning Techniques for Churn Prediction and Factor
Identification in Telecom Sector," IEEE Access, vol. 7, pp.
60134–60149, 2019.
39. Z. B. M. Saghir, S. Bashir, and F. H. Khan, "Churn Prediction
using Neural Network based Individual and Ensemble Models,"
presented at the 16th International Bhurban Conference on
Applied Sciences and Technology (IBCAST), 2019.
40. X. Cai, Shang, J., Jin, Z., Liu, F., Qiang, B., Xie, W., "DBGE:
Employee turnover prediction based on dynamic bipartite graph
embedding," IEEE Access, vol. 8, pp. 10390–10402, 2020.
41. M. K. M. P. Lalwani, J. S. Chadha, and P. Sethi, "Customer churn
prediction system: a machine learning approach," Computing, pp.
1-24, 2021.
42. S. H. A. Khodadadi, E. Pajouheshgar, F. Mansouri, and H. R.
Rabiee, “ChOracle: A unified statistical framework for churn
prediction,” IEEE Trans. Knowl. Data Eng., pp. 1–1, 2020,
"ChOracle: A unified statistical framework for churn prediction,"
IEEE Trans. Knowl. Data Eng., pp. 1-12, 2020.
43. M. J. P. K. Jain, and R. Pamula, "Explaining and predicting
employees’ attrition: a machine learning approach," SN Appl.
Sci., vol. 2, pp. 1-11, 2020.
44. A. M. D. Mukhopadhyay, A. Nandanwar, and S. Sakshi, "An
approach to mitigate the risk of customer churn using machine
learning algorithms," in Machine Learning for Predictive Analysis,
Singapore, 2021, pp. 133–142.

Journal of Information Communication Technologies and Robotic Applications


http://www.jictra.com.pk/index.php/jictra, pISSN: 2523-5729, eISSN: 2523-5739
JICT R A 2 021 9

You might also like