IBM Employee Attrition Analysis

Shenghuan Yang Md Tariqul Islam

Jiangxi University of Finance and Economics Syracuse University

ABSTRACT 2 RELATED WORK

In this paper, we analyzed the dataset IBM Employee Attri- Alao et al.[3] used decision tree models and rule-sets to develop
tion to find the main reasons why employees choose to re- a predictive model that was used to predict new cases of em-
sign. Firstly, we utilized the correlation matrix to see some ployee attrition. Alduayj et al.[4]utilized support victor machine
features that were not significantly correlated with other at- (SVM) with several kernel functions, random forest and Knearest
tributes and removed them from our dataset. Secondly, we neighbour (KNN) to predict employee attrition based on their fea-
selected important features by exploiting Random Forest, tures and found quadratic SVM scored the highest results. Frye
arXiv:2012.01286v6 [cs.CY] 1 Jan 2021

finding monthlyincome, age, and the number of companies et al.[5]applied Principal Component Analysis and classification
worked significantly impacted employee attrition. Next, we methods K-Nearest Neighbors and Random Forest, finding that
also classified people into two clusters by using K-means Logistic Regression predicts employee quits with the highest accu-
Clustering.Finally, we performed binary logistic regression racy. Yadavet et al.[6] provided a framework for predicting the em-
quantitative analysis: the attrition of people who traveled ployee churn by analyzing the employee’s precise behaviors and at-
frequently was 2.4 times higher than that of people who tributes using classification techniques. Srivastava et al.[7]provided
rarely traveled. And we also found that employees who work a framework for predicting the employee churn by analyzing the
in Human Resource have a higher tendency to leave. employee’s precise behaviors and attributes using classification
techniques. El-Rayes et al.[8] presented a framework for predicting
the employee attrition with respect to voluntary termination em-
ploying predictive analytics. Setiawan et al.[9] found that eleven
1 INTRODUCTION variables that have a significant impact on employee attrition.
Employee attrition is defined as the natural process by which em-
ployees leave the workforce – for example, through resignation 3 DATA AND METHODOLOGY
for personal reasons or retirement – and are not immediately re-
placed[1]. Employee turnover is regarded as the key issue for all Employee attrition is the internal data of the company, which is
organizations these days, because of its adverse effects on work- difficult to obtain, and some data has a certain degree of confiden-
place productivity, and accomplishing organizational objectives on tiality, therefore our paper used the data set disclosed by kaggle.
time [2]. In order for an organization to continually have a higher The sample size of the data set is 1471, there are 34 feature vari-
competitive advantage over its competition, it should make it a ables,mainly divided into three types of variables: personal basic
duty to minimize employee attrition Therefore, for the better devel- information, work experience, attendance rate.This paper explored
opment of corporation, it is essential for the leader of companies to the relationship between employee’s characteristics and employee
know the main reasons why their employees choose to leave the attrition, found Whether characteristics have a great influence on
company,then take relevant measures to improve their company’s employee attrition. In addition, we used machine learning algo-
productivity, overall workflow and business performance. rithms to select important features that influenced the employee
Objectives. In this paper, we aim to select the main causes that attrition, and predicted the it. In this paper, we exploited three ma-
contribute to an employee’s decision to leave a company, and to chine learning algorithms: Decision Tree, and Logistic Regression
be able to predict whether a particular employee will leave the and k-means clustering.
company by utilizing machine learning models.
Contributions. Following are the main contributions of this 3.1 Random Forest
paper: Random forest is an learning method used for classification, regres-
• We select the main factors affecting the employee attrition sion and other tasks, which combines multiple decision trees to
by using Random Forest, and classify which types of people select the best result. Random forest corrects the habit of decision
are more likely to quit by utilizing the K-means Clustering trees that rely too much on the training set and improves the accu-
• We represent a given reality in terms of a numerical value racy of the model.First, there are randomly selected sub-data sets
to compare the employee attrition in different categories by that are replaced from the original data[10]. The elements between
utilizing quantitative analysis. the sub-datasets may have the same elements, and the content of
each sub-data set is 1 different. Then use the sub-dataset to con-
The rest of the paper is described below. We present some related struct the sub-decision tree, each decision tree will output a result,
work in Section 2. We interpret data and introduce methodologies you can vote through the output result of the sub-decision tree,
in Section 3. We process the original data set and remove some vari- and finally get the output result. As shown in Figure 1, the data
ables which are not very correlated with other features in section set extracts 4 sub-datasets to construct 4 sub-decision trees, 3 trees
4. We implement our machine learning model in Section 5. Finally, voted as A, one sub-decision tree voted as B, and the final output is
we draw a conclusion in section 6. A.
 Shenghuan Yang and Md Tariqul Islam

are coefficients, which represent the influence of the variable on

the dependent variable. Whether it is positive or negative, it needs
to be explained in conjunction with the corresponding regression
coefficient value. If the regression coefficient value is greater than
0, it means a positive influence; otherwise, it means a negative
influence for the dependent variable.

3.3 k-means Clustering

The K-means Clustering algorithm is the most commonly used
clustering algorithm. The main idea is: given K values and K initial
cluster centers, assign each point to the nearest cluster center, after
all the points are allocated, the center point of the cluster is recalcu-
lated based on all the points in the cluster (take the average value),
updates the center point of the cluster Steps until the center point
of the cluster changes very little [13]. The objective function is:
𝑘 ∑︁
𝑛
Figure 1: Decision-making process of random forest ( 𝑗)
∑︁
𝐽 = ||𝑥𝑖 − 𝑐 𝑗 || 2 (2)
𝑗=1 𝑖=1
Random forest is one of the most popular machine learning ( 𝑗)
Where ||𝑥𝑖 − 𝑐 𝑗 || is the Euclidean distance between 𝑥𝑖 and
( 𝑗)
algorithms. Random forest provides a unique combination of pre-
𝑐 𝑗 , n is the number of data points in 𝑗 𝑡ℎ cluster, k is the number
diction accuracy and model interpretability among popular machine
of cluster centers. Algorithmic steps for k-means clustering: 𝑋 =
learning methods. The random sampling and ensemble strategies
𝑥 1, 𝑥 2, 𝑥 3, ..., 𝑥𝑛 is the set of data points, and 𝑐 = 𝑐 1, 𝑐 2, ..., 𝑐𝑘 is the
utilized in RF enable it to achieve accurate predictions as well as
set of centers.
better generalizations [11]. In addition, it has a strong explanatory
nature. It can directly calculate the importance of each variable. (1) Randomly select ’k’ cluster centers
In other words, it is easy to calculate how much each variable (2) Calculate the distance between each data point and cluster
contributes to decision-making. In our research, we build Random centers
Forest model based on Employee Attrition Features.There are 34 (3) Assign the data point to the cluster center whose distance
employee attributes in the data set, we select randomly k(k<34) from the cluster center is minimum of all the cluster centers
employee attributes to build a decision tree, and create 100 random (4) Recalculate the new cluster center using:
sub-samples of our dataset with replacement. Each node of each 𝑛
decision tree will be split according to predictor variables so that 1 ∑︁
𝑐𝑗 = 𝑥𝑖 (3)
child nodes are more “pure” (i.e., Gender) in terms of the outcome 𝑛 𝑖=1
variable. If the feature of the next child node has appeared in the (5) Recalculate the distance between each data point and new
parent node, the second child node will stop splitting. In the end, obtained cluster centers
each decision tree outputs a result, and the result with the most (6) If no data point was reassigned then stop, otherwise repeat
votes is the final result. from step (3)
The process of K-means Clustering is shown in Figure 2.
3.2 Binary Logistic Regression
Logistic regression analysis can be used to estimate the probability
of an event, and it can also analyze the influencing factors of the
problem. In medical research, Logistic regression is often used to
analyze the risk factors of a certain disease[12]. Such as analyzing
whether age, smoking, drinking, and diet are risk factors for dia-
betes. In questionnaire research, Logistic regression is often used to
analyze non-scale questions, such as taking the basic background
information of the sample as X and purchasing intention as Y to an-
alyze whether gender, age, and families’ socioeconomic conditions
affect purchase intention. Among them, binary logistic regression
analysis is used most frequently.

𝑝 (𝑦 = 1)
𝑙𝑜𝑔𝑖𝑡 ( 𝑝) = ln = 𝛽 0 + 𝛽 1𝑥 1 + ... + 𝛽𝑛 𝑥𝑛 (1)
1 − 𝑝 (𝑦 = 1)
Here, employee attrition as the dependent variable y, 0 represents
No, 1 represents Yes. Gender, EducationLevel, OverTime, JobLevel
and so on as predictor variables, n=34, 𝑥 1 ,𝑥 2 ,..., 𝑥 34 . 𝛽 0 , 𝛽 1 ,..., 𝛽𝑛 Figure 2: Clustering process of K-means Clustering
IBM Employee Attrition Analysis

4 DATA PROCESS employee attrition, and ignore the variables that are not significant
In our studies, there are 34 variables, however, some features just in explaining such behavior.
have one data level that do not make sense for our research, such as
EmployeeCount, Over18 and StandardHours, and employee number
does not have meaning in analyzing resulting so we also deleted
these features. In addition, some features are not very correlated
with other attributes, we need to remove them from our dataset to
improve computing efficiency. We built correlation matrix, which
is a table showing correlation coefficients between variables. Each
cell in the table shows the correlation between two variables.

Figure 4: Important Features

From Figure 4, Monthly Income, Age, DisatnceFromHome are

top 3 important features to indicate whether the employee has a
tendency to leave, while marital status married, and female aged 40-
50 are less likely to go. Indeed, income is the main cause that people
choose to leave the company, which is highly related to people’s life
Figure 3: Correlation Matrix quality. More affluent people have more disposable income and can
more easily afford expensive service (such as medical care) and a
healthy lifestyle—benefits.So people want to have more money that
We can see from Figure 3, daily rate, hourly rate and monthly
they intend to leave the current company and find other companies
rate are barely correlated with other attributes, while Monthly
that offer high salaries. And also young people are more likely to
Income and job level (0.95), Job level and total working years(0.78),
want to try different jobs and finally find a job that suits them when
Monthly Income and total working years(0.77) are highly correlated.
they just graduated from universities. Whereas older people prefer
Therefore we delete daily rate, hourly rate and monthly rate from
a stable life because they already have own families and children,
dataset.
so it is not easy to leave the job. In addition, the distance between
the company and home is also the main reason why people quit the
5 MODEL CONSTRUCTION
current job, they would waste much time commuting. There’s mass
In this part, we build machine learning models to select impor- of social science and public health research on the negative effects of
tant features that influence the employee attrition and classify the commuting on personal and societal well-being. Longer commutes
features to help us understand the main reason why people left are linked with increased rates of obesity, high cholesterol, high
the IBM company.And we will construct the model to predict the blood pressure, back and neck pain, divorce, depression and death.
employee attrition according to the given data. Here, we need to train and test dataset to predict the employee
attrition according to the features. We splited the data set into
5.1 Feature Selection two sets: a training set and a testing set (80% for training, and
We all may have faced this problem of identifying the related fea- 20% for testing). To better reflect the real situation and avoid the
tures from a set of data and removing the irrelevant or less impor- influence of extreme values, we predicted the employee attrition for
tant features with do not contribute much to our target variable 100 times by utilizing RandomForest Classifier, and recorded the
in order to achieve better accuracy for our model and.Therefore, average accuracy of the test set. Table 1 shows the average accuracy
it is better for us to select important features that influence the is 0.84561, which means our model fits well.
 Shenghuan Yang and Md Tariqul Islam

5.2 Classification of people job chance. What is more, job satisfaction is also the main cause
To distinguish which types of people are more likely to resign, we influencing the employee attrition rate.Intention to stay on the job
use K-means clustering to divide the dataset into two categories. is clearly correlated with job satisfaction in such aspects as edu-
The first type is prone to leave, and the other is less likely to quit. cational system and environment, income and welfare, leadership
We can see from Table 2 (complete form in appendix table 4),clus- and administration.[14].
ter set 0 represents low attrition, cluster set 1 means high attrition,
and older people, high job level, high job satisfaction,high monthly 5.3 Logistics Regression
income, more number of companies worked and so forth are less In this section, we exploited the binary logistics regression to pre-
likely to leave.These finding are in line with people’s behavior in the dict the relationship between predictors (employee characteristics)
real world and previous accounts in feature selection of Random- and a predicted variable (employee attrition) where the dependent
Forest. In last section, we illustrated the relationship between age, variable is binary (NO:0, YES:1). And we will compare the differ-
monthly income, distance from the company and home, this section, ences between each category to help us understand which type of
we also find that the number of companies employees worked are person is more likely to quit (complete regression table can be seen
related to the probability of leaving the corporation. People who in appendix table 5).
have worked in 3-4 companies are less likely to quit because by this
time, they have roughly found the direction of employment, and Table 3: Regression Result
people who have more than four companies indicate that they are
unstable and often change jobs. In addition, people who are in the OR(95% CI) P-value
higher the job level, enjoy the higher salary and respect , so they are Travel_Rarely 1
less likely to leave. While those employees in lower-level positions NonTravel 0.361
are often not satisfied with the status quo and want to seek better Travel_Frequently 2.411 <0.05
Male 1
Table 1: Prediction Accuracy of Random Forest Female 0.659 <0.05
Sales Representative 1
Healthcare 0.160
Times Accuracy
Human Resource 4.060
1 0.85714 Laboratory 0.556
2 0.83673 Manager 0.347
3 0.84353 Manufacturing 0.200
4 0.84693 Research 0.200
... ... Sales Executive 0.484 <0.05
100 0.85374 Single 1
Average Accuracy 0.84561 Divorced 0.304
Married 0.427 <0.05
Table 2: Clustering OverTime 0.138 <0.05

Cluster Type 0 1 From Table 3, we can see, taking people who rarely travel as the
standard 1, employees who travel frequently are 0.361 times more
Age 44.215152 34.813158 than those to leave the company whereas people who never travel
Attrition 0.1 0.178947 are more than twice. And Women are 0.659 times more than men
DistanceFromHome 9.072727 9.227193 to go. As for JobRole, taking Sales Representative as the standard
Education 3.039394 2.876316 1, employee who who work about Research Science, Laboratory,
EnvironmentSatisfaction 2.693939 2.729825 Manufacturing, Healthcare are not likely to quit, while Human-
JobInvolvement 2.690909 2.741228 Resource Department has high employee attrition. Then single
JobLevel 3.684848 1.594737 people are more likely to leave their jobs than married and divorced
JobSatisfaction 2.709091 2.734211 people.Finally,not surprised, people who usually work overtime
MonthlyIncome 14060.49394 4315.215789 would like to leave the company. To evaluate the logistics regres-
NumCompaniesWorked 3.342424 2.505263 sion model, we need to test the accuracy of model in Python, the
OverTime 0.290909 0.280702 test result is 0.8843, which means our model fitting is well.
... ... ...
... ... ... 6 CONCLUSION
PercentSalaryHike 15.066667 15.250877
According the above model results, we can know that our finding
PerformanceRating 3.148485 3.155263
are in line with people’s behavior in the real world and previous
RelationshipSatisfaction 2.781818 2.692105
studies other scholar did. We utilized Random Forest and K-means
StockOptionLevel 0.80303 0.791228
Clustering to select important features that had obvious impact on
TotalWorkingYears 21.072727 8.444737
the employee attrition. Firstly, according to Random Forest results,
IBM Employee Attrition Analysis

monthlyincome, age, the number of companies worked are the [12] Meng X H, Huang Y X, Rao D P, et al. Comparison of three data mining models
main reasons why people choose to resign. Then we found older for predicting diabetes or prediabetes by risk factors[J]. The Kaohsiung journal
of medical sciences, 2013, 29(2): 93-99.
people, high job level, high job satisfaction,high monthly income, [13] Pham D T, Dimov S S, Nguyen C D. Selection of K in K-means clustering[J]. Pro-
more number of companies worked, these kinds of people are not ceedings of the Institution of Mechanical Engineers, Part C: Journal of Mechanical
Engineering Science, 2005, 219(1): 103-119.
likely to go based on the clustering result of K-means Clustering. [14] Weiqi, Chen. "The structure of secondary school teacher job satisfaction and its
However, different people have various intention, we need to do relationship with attrition and work enthusiasm." Chinese Education& Society
further and detailed research to find qualitative findings by using 40.5 (2007): 17-31.
qualitative analysis. So we exploited the binary logistics regression
to compare the difference between people. Our study found that
females’ attrition was 0.659 times than that of males, married and
divorced people were 0.427 and 0.304 times than people who were
single, respectively. Besides, the attrition of people who traveled
frequently was 2.4 times higher than that of people who rarely
traveled. And we also found that employees who work in Human
Resource have a higher tendency to leave. Finally, there are other
interesting findings in our study:in terms of number of companies
worked, people who worked in 2 - 4 companies are less likely to
leave, the female attrition rate is less than male after working for
six companies , and people who earned Doctor’s Degree are almost
always having the lowest attrition rate.
To evaluate the model performance, we trained and tested the
dataset to predict the employee attrition, split it into two parts(80%
for training, 20% for testing ), and recorded the test set’s accuracy.
Random Forest and Logistics Regression accuracy were 0.8456 and
0.8843, respectively, which meant Logistics Regression fitted better
and was more suitable for prediction in our dataset.
We also want to make some suggestions to the company through
this research, hoping that they will care more about their employ-
ees and improve their job satisfaction. Simultaneously, they must
pay more attention to human resources employees because they
have very low job satisfaction. Besides, the company should allow
employees to have enough time to rest and spend time with their
families.There is a general belief that employees who take regular
breaks are more productive.

REFERENCES
[1] Budhwar P S, Bhatnagar J. Talent management strategy of employee engagement
in Indian ITES employees: key to retention[J]. Employee relations, 2007.
[2] Jain, Rachna, and Anand Nayyar. Predicting employee attrition using xgboost
machine learning approach. 2018 International Conference on System Modeling
& Advancement in Research Trends (SMART). IEEE, 2018.
[3] Alao, D. A. B. A., and A. B. Adeyemo. "Analyzing employee attrition using decision
tree algorithms." Computing, Information Systems, Development Informatics
and Allied Research Journal 4.1 (2013): 17-28.
[4] Alduay j, Sarah S., and Kashif Rajpoot. "Predicting employee attrition using
machine learning." 2018 International Conference on Innovations in Information
Technology (IIT). IEEE, 2018.
[5] Frye, Alex, et al. "Employee Attrition: What Makes an Employee Quit?." SMU
Data Science Review 1.1 (2018): 9.
[6] Yadav, Sandeep, Aman Jain, and Deepti Singh. "Early Prediction of Employee
Attrition using Data Mining Techniques." 2018 IEEE 8th International Advance
Computing Conference (IACC). IEEE, 2018.
[7] Srivastava, Devesh Kumar, and Priyanka Nair. "Employee attrition analysis using
predictive techniques." International Conference on Information and Communi-
cation Technology for Intelligent Systems. Springer, Cham, 2017.
[8] El-Rayes, Nesreen, et al. "Predicting employee attrition using tree-based models."
International Journal of Organizational Analysis (2020).
[9] Setiawan, I., et al. "HR analytics: Employee attrition analysis using logistic re-
gression." IOP Conference Series: Materials Science and Engineering. Vol. 830.
No. 3. IOP Publishing, 2020.
[10] Pal M. Random forest classifier for remote sensing classification[J]. International
journal of remote sensing, 2005, 26(1): 217-222.
[11] Qi, Yanjun. "Random forest for bioinformatics." Ensemble machine learning.
Springer, Boston, MA, 2012. 307-323.
 Shenghuan Yang and Md Tariqul Islam

Table 4: Clustering Result

Cluster Type 0 1
Age 44.215152 34.813158
Attrition 0.1 0.178947
DistanceFromHome 9.072727 9.227193
Education 3.039394 2.876316
EnvironmentSatisfaction 2.693939 2.729825
JobInvolvement 2.690909 2.741228
JobLevel 3.684848 1.594737
JobSatisfaction 2.709091 2.734211
MonthlyIncome 14060.49394 4315.215789
NumCompaniesWorked 3.342424 2.505263
OverTime 0.290909 0.280702
PercentSalaryHike 15.066667 15.250877
PerformanceRating 3.148485 3.155263
RelationshipSatisfaction 2.781818 2.692105
StockOptionLevel 0.80303 0.791228
TotalWorkingYears 21.072727 8.444737
TrainingTimesLastYear 2.79697 2.8
WorkLifeBalance 2.781818 2.755263
YearsAtCompany 11.927273 5.584211
YearsInCurrentRole 6.133333 3.67807
YearsSinceLastPromotion 4.109091 1.631579
YearsWithCurrManager 5.821212 3.631579
Male 0.581818 0.605263
BusinessTravelNon-Travel 0.087879 0.10614
BusinessTravelTravelFrequently 0.187879 0.188596
BusinessTravelTravelRarely 0.724242 0.705263
DepartmentHuman Resources 0.045455 0.042105
DepartmentResearch & Development 0.642424 0.657018
DepartmentSales 0.312121 0.300877
EducationFieldHuman Resources 0.021212 0.017544
EducationFieldLife Sciences 0.39697 0.416667
EducationFieldMarketing 0.130303 0.101754
EducationFieldMedical 0.324242 0.313158
EducationFieldOther 0.045455 0.058772
EducationFieldTechnical Degree 0.081818 0.092105
JobRoleHealthcare Representative 0.112121 0.082456
JobRoleHuman Resources 0.012121 0.042105
JobRoleLaboratory Technician 0 0.227193
JobRoleManager 0.309091 0
JobRoleManufacturing Director 0.121212 0.092105
JobRoleResearch Director 0.242424 0
JobRoleResearch Scientist 0.00303 0.255263
JobRoleSales Executive 0.2 0.22807
JobRoleSales Representative 0 0.072807
MaritalStatusDivorced 0.251515 0.214035
MaritalStatusMarried 0.506061 0.44386
MaritalStatusSingle 0.242424 0.342105
IBM Employee Attrition Analysis

Table 5: Logistics Regression Result

95% C.I.for EXP(B)

B S.E. Wald df Sig. Exp(B) Lower Upper
Age -0.031 0.014 5.285 1 0.02∗∗ 0.969 0.944 0.995
TravelRarely 27.561 2 0.00∗∗
NonTravel -1.020 0.381 7.175 1 0.02∗∗ 0.361 0.171 0.761
TravelFrequently 0.880 0.211 17.323 1 0.04∗∗ 2.411 1.593 3.649
DailyRate 0.000 0.000 1.645 1 0.200 1.000 0.999 1.000
Sales Department 0.010 2 0.995
HumanResource Department -18.223 10790.424 0.000 1 0.999 0.000 0.000
ResearchDepartment 0.104 1.043 0.010 1 0.921 1.109 0.144 8.566
DistanceFromHome 0.046 0.011 17.837 1 0.00∗∗ 1.047 1.025 1.069
Education 0.007 0.088 0.006 1 0.937 1.007 0.848 1.196
Human Re 13.256 5 0.02∗∗
Life Science -0.189 0.823 0.053 1 0.818 0.828 0.165 4.152
Marketing -0.926 0.304 9.279 1 0.00∗∗ 0.396 0.218 0.719
Medical -0.525 0.394 1.777 1 0.182 0.591 0.273 1.280
Other -1.036 0.312 11.029 1 0.00∗∗ 0.355 0.193 0.654
Technical -0.983 0.475 4.276 1 0.00∗∗ 0.374 0.147 0.950
EmployeeNumber 0.000 0.000 0.888 1 0.346 1.000 1.000 1.000
EnvironmentSatisfaction -0.431 0.083 26.988 1 0.00∗∗ 0.650 0.553 0.765
Gender(1) -0.417 0.184 5.098 1 0.00∗∗ 0.659 0.459 0.947
HourlyRate 0.001 0.004 0.087 1 0.768 1.001 0.993 1.010
JobInvolvement -0.528 0.123 18.417 1 0.00∗∗ 0.590 0.463 0.751
JobLevel -0.128 0.317 0.162 1 0.688 0.880 0.472 1.640
Sales Representative 20.855 7 0.00∗∗
Healthcare -1.835 1.164 2.484 1 0.115 0.160 0.016 1.563
Human Resource 17.601 10790.424 0.000 1 0.999 44066714.479 0.000
Laboratory -0.588 1.090 0.291 1 0.590 0.556 0.066 4.702
Manager -1.058 0.985 1.154 1 0.283 0.347 0.050 2.392
Manufacturing -1.609 1.162 1.919 1 0.166 0.200 0.021 1.950
Research -1.611 1.097 2.156 1 0.142 0.200 0.023 1.715
Sales Executive -0.726 0.385 3.555 1 0.059 0.484 0.228 1.029
JobSatisfaction -0.414 0.081 25.891 1 0.00∗∗ 0.661 0.563 0.775
Single 14.274 2 0.00∗∗
Divorced -1.191 0.346 11.885 1 0.00∗∗ 0.304 0.154 0.598
Married -0.852 0.251 11.525 1 0.00∗∗ 0.427 0.261 0.698
MonthlyIncome 0.000 0.000 0.292 1 0.589 1.000 1.000 1.000
MonthlyRate 0.000 0.000 0.138 1 0.710 1.000 1.000 1.000
NumCompaniesWorked 0.196 0.039 25.654 1 0.00∗∗ 1.216 1.128 1.312
OverTime(1) -1.984 0.194 104.913 1 0.00∗∗ 0.138 0.094 0.201
PercentSalaryHike -0.021 0.039 0.287 1 0.592 0.979 0.907 1.058
PerformanceRating 0.096 0.398 0.058 1 0.810 1.100 0.504 2.402
RelationshipSatisfaction -0.270 0.083 10.591 1 0.00∗∗ 0.763 0.649 0.898
StockOptionLevel -0.186 0.158 1.387 1 0.239 0.830 0.609 1.131
TotalWorkingYears -0.057 0.029 3.772 1 0.052 0.945 0.892 1.001
TrainingTimesLastYear -0.193 0.073 6.976 1 0.00∗∗ 0.825 0.715 0.951
WorkLifeBalance -0.377 0.124 9.298 1 0.00∗∗ 0.686 0.538 0.874
YearsAtCompany 0.098 0.039 6.348 1 0.00∗∗ 1.103 1.022 1.191
YearsInCurrentRole -0.148 0.045 10.701 1 0.00∗∗ 0.862 0.789 0.942
YearsSinceLastPromotion 0.175 0.042 16.934 1 0.00∗∗ 1.191 1.096 1.294
YearsWithCurrManager -0.138 0.047 8.575 1 0.00∗∗ 0.871 0.794 0.955
Constant 9.308 1.383 45.275 1 0.00∗∗ 11022.507
**significant at p < 0.05