VIETNAM NATIONAL UNIVERSITY, HANOI INTERNATIONAL SCHOOL GRADUATION PROJECT PROJECT NAME EMPLOYEE CHURN PREDICTION Student’s name: Nguyễn Phương Ly Hanoi - Year 2021 1 VIETNAM NATIONAL UNIVERSITY, HANOI INTERNATIONAL SCHOOL GRADUATION PROJECT PROJECT NAME EMPLOYEE CHURN PREDICTION SUPERVISOR: DR.TRAN THI OANH STUDENT: NGUYEN PHUONG LY STUDENT ID: 17071362 MAJOR: MANAGEMENT INFORMATION SYSTEM Hanoi - Year 2021 2 LETTER OF DECLARATION I hereby declare that the Graduation Project EMPLOYEE CHURN PREDICTION is the results of my own research and has never been published in any work of others. During the implementation process of this project, I have seriously taken research ethics; all findings of this project are results of my own research and surveys; all references in this project are clearly cited according to regulations. I take full responsivity for the fidelity of the number and data and other contents of my graduation project. Hanoi, 08 /06/2021 Student (Signature and Full name) LY Nguyễn Phương Ly 3 ACKNOWLEDGEMENT First and foremost, I would like to express my gratitude to my supervisors, Dr Tran Thi Oanh, for their invaluable advice, unwavering support, and patience throughout my graduation thesis.
Their vast knowledge and wealth of experience have inspired me throughout my academic research and daily life. My thanks go to the Faculty of science and technology department for providing me with the funding to pursue my studies at the science and technology department, University of International School – Hanoi National University. My thanks also go to my family and friends for their motivation and help throughout my studies. 4 Table of Contents LETTER OF DECLARATION .4 List of figures.8 List of table.
10 Chapter 1: Introduction of Employee Churn Prediction. Introduction employee churn prediction. The meaning of problem. 15 Chapter 2: Machine learning methods used in classification.
Decision Tree Algorithm. Support vector machine (SVM). Extreme Gradient Boosting (Xgboost). 33 Chapter 3: Data Exploration.
How many candidates left the company?. Which gender has the most number of people leaving the company?. Do more experienced employees look for a new job?. Do employees in smaller businesses yearn for new opportunities?.
Which education level has the most number of people looking for new job?. Does amount of training hour impact on decision to leave company?. Does City development index impact on decision to leave company?. What is the relationship between training hour, city development index and target? 40 Chapter 4: Predicting employee churn using machine learning methods.
54 Chapter 5: Result and Evaluation. Experimental result by using Individual classifier. Experimental result by using Ensemble learning. Comparing between individual classifier and ensemble learning.
Experimental result using SMOTE. 69 7 List of figures Figure 1.1: RELATED WORK ON TURNOVER PREDICTION (source: Punnoose, R. Prediction of Employee Turnover in Organizations using Machine Learning Algorithms. International Journal of Advanced Research in Artificial Intelligence, 5(9).2: Architecture of the employee churn prediction .3: Formula of monthly turnover (%) .1: Classification problem modeling ‘Yes’/‘No’ response based on three features.
(Source: https://bradleyboehmke.io/HOML/process.3: Structure of Decision tree (source: https://www.com/community/tutorials/decision-tree-classification-python).4: How does decision tree work (source: https://www.com/community/tutorials/decision-tree-classification-python).5: KNN Algorithm (source: https://www.com/community/tutorials/k- nearest-neighbor-classification-scikit-learn) .6: Steps of KNN (source: https://www.com/community/tutorials/k- nearest-neighbor-classification-scikit-learn.7: One hidden layer MLP. (source: Scikit-learn. Neural network models (supervised) — scikit-learn 0. [online] Available at: <https://scikit- learn.org/stable/modules/neural_networks_supervised.9: Architecture of random forest (source: https://www.com/community/tutorials/random-forests-classifier-python) .10: The sequence ensemble methods ( source: https://www.com/community/tutorials/xgboost-in-python) .1: How many candidates leave the company? (%) .2: Distribution of gender and target by gender .3: Distribution of experiences and target by experiences .4: Distribution of company size by class 1 ('Leaving') .5: Distribution of education level by class 1 ('Leaving') .6: Box plot of target by training hour .7: Box plot of target by City development index .8: relationship between training hour, city development index and target .1: Classification problem modeling ‘Not leaving’/‘leaving’ response based on 12 features .3 : Distribution of training hour and city development index .4: Distribution of target .5: Example of dummy encoding .6: Total number of missing value (%) .7: Filling missing values using fillna() .8: Check null values after using fillna() .1: Weighted F1 score of Decision tree, KNN, Logistic regression, SVM, MLP classifier, Random forest, soft voting classifier, hard voting classifier and XGBoost .2: Weighted F1 score of Class 1 and Class 0 without SMOTE and with SMOTE .3: Weighted F1 score of selected features model and full features models.
66 List of table Table 5-1: Classification report of logistic regression (%). 57 Table 5-2: Classification report of Decision tree (%). 58 Table 5-3: Classification report of KNN (%). 58 Table 5-4: Classification report of MLP classifier.
59 Table 5-5: Classification report of SVM (%). 59 Table 5-6: Classification report of Random Forest (%). 60 Table 5-7: Classification report of XBGoost (%). 60 Table 5-8: Classification report of Soft Voting classifier (%).
61 Table 5-9: Classification report of Hard Voting classifier (%). 61 Table 10: Precision, Recall and F1-score of XGBoost model without SMOTE and with SMOTE (%). 63 9 Abstract In the SHRM/Globoforce survey Using Recognition and Other Workplace Efforts to Engage Employees, 47 percent of HR professionals listed retention/turnover as the top workforce management challenge. Revolving workforces frequently result in higher training expenses, irregular production, low morale, and, as a result, lower or limited profitability.
Therefore, it is necessary to focus on reducing turnover. In this study, we target to building a prediction model to predict employee churn using machine learning-individual classifier methods such as Decision Tree, Logistic Regression, SVM, KNN, MLP classifier and ensemble learning such as XGBoost, Random Forest, and Voting classifier. To evaluate the effectiveness of the proposed model, we perform extensive experiments on a public dataset of HR Analytics: Job Change of Data Scientist dataset. To understand more about the dataset before using it, we also explore the data by using different descriptive methods such as visualization, some statistical methods, etc.
The findings will enable businesses to estimate their employees' churn rates and, as a result, lower their human resource costs. 10 Chapter 1: Introduction of Employee Churn Prediction 1. Introduction employee churn prediction There are numerous case studies available on customer churn that predicts whom and when a customer will stop buying. Employee churn or employee turnover is similar to customer churn in many ways.
It is primarily concerned with the employee rather than the customer. Here, employee churn refers to people leaving their jobs in an organization. Employee churn is a significant issue for organizations, particularly when trained, technical and key employees leave for a better opportunity with a competitor organization. Organizations can design successful policies and initiatives for employee retention if they understand why their employees are likely to leave.
Induction crises, natural wastage, and retirement are the three main reasons for an organization's staff loss. Beside, high job turnover has been linked to a lack of professional possibilities, challenges, unhappiness with the job scope, and disagreement with management. (1) Voluntary and involuntary employee turnover are the two types of employee turnover. Employees initiate voluntary turnover; for example, a worker quits and takes another position.
The organization causes involuntary turnover; for example, a firm may fire an employee owing to poor performance or organizational restructuring. Employee turnover is also uncontrollable in the case of involuntary turnover, such as retirement, firing, or death. The concept of turnover intent can be used to forecast (and thus control) voluntary turnover. The most essential issue that industries should consider is voluntary turnover.
By searching databases for hidden patterns, data mining technologies can forecast future trends and behaviors, allowing businesses to make proactive knowledge-driven decisions and answer questions that were previously too time-consuming to address. Some data mining techniques include Decision Trees, Logistic analysis, ANN, Rule induction, Clustering, Association rule and correlation… (1) The findings of the literature review are summarized in Figure 1 below: Figure 1.1: RELATED WORK ON TURNOVER PREDICTION (source: Punnoose, R. Prediction of Employee Turnover in Organizations using Machine Learning Algorithms. International Journal of Advanced Research in Artificial Intelligence, 5(9).) 12 In this study, the topic of employee turnover is described, as well as the important machine learning techniques that have been utilized to solve it.
Employee churn prediction use variables that related to employee, for example age, gender, ethnicity, education, and marital status, tenure, pay, overall job satisfaction, and employee’s perceptions of fairness… as input to gain "churn" or "not churn" output. Let's look at the figure below show architecture of the employee churn prediction: Figure 1.2: Architecture of the employee churn prediction Training and career development are two of the most important factors millennials consider when determining the attractiveness of a job or company as a place to work. Furthermore, 46 percent of employees believe that their company's training courses make them more likely to stay. But to have good training courses, companies need to consider information of candidates who signup for their training.
Then based on this information they might predict which of these candidates are really wants to work for the company after training or looking for a new employment. This step is extremely important in categorization of candidates to have appropriate planning course for each ones. Moreover, it helps reduce time, improve the quality of training, and especially reduce cost. As you also know, when an employee leaves, the companies losing more than the money their invested in training and there are a few 13 of expensive they could face such as: paying a new employees 15-35 percent more than their former employees; a recruiter's fee of 20-30 percent of the new employee's salary and ironically enough, payback of sign-on bonus and training fees to the new employees previous employer.
Therefore, It is very useful to companies for applying machine learning techniques to predict employee churn, which helps them in taking necessary actions. Motivation HR analytics often answer the questions related to employee turnover rate. The churn rate is usually calculated as the percentage of employees leaving the company over some specified time period. For example calculate the monthly employee turnover rate, all you need is three numbers: the numbers of active employees at the beginning (B) and end (E) of the month and the number of employees who left (L) during that month.3: Formula of monthly turnover (%) It is a way to compare your company’s turnover rate with the average rate within your industry or support predict turnover trends when collect data from different periods of time and only show the impact of churn or turnover in the ‘aggregate’.
To predict exactly ‘who’ or ‘which employees’ exactly may be at high risk of leaving the organization or focus individual rather than ‘aggregate’, you should to 14 apply another approach. In this study, I use machine learning techniques to predict employee churn. There are several studies on employee churn prediction using machine learning methods. These study often use information employee such as age, gender, role and work conditions, satisfaction level, promotion rate ,salary and so on.
It seems to be churn prediction churn of best and experienced or former employees rather than focusing on fresher candidates churn. This reason why I want to pay attention to fresher candidates churn prediction. It is very useful for companies to know which candidates are really wants to work for the company after training or which are looking for a new employment. This result helps reduce the cost and time as well as improve the quality of training or planning the courses and categorization of candidates.
The meaning of problem Employee turnover is expensive from the view of the organization. Understanding why and when employees are most likely to leave bring many benefits for organization. This make me more interested in investigating the employee turnover prediction.