Group14 proposal

Proposal

Poster

Application

Research Paper

Motivation and Objectives

What HRs are facing is how well they can retain their talent and better control the employee turnover. Among all the employee-related problems for businesses around the world, attrition is one of the significant problems regardless of the changes in the external working environment. Because of today’s competitive job market, high employee attrition is one of the most pressing issues businesses face. By 2023, voluntary employee turnover is expected to rise to nearly 30%. Some major impacts on employee attrition, including lack of employee continuity and possible high costs involved in the induction and training of new staff will result in issues of organizational productivity.

Hence, it really matters what will affect the employee and analyze what we can do to reduce as well as adjust the structure of personnel in order to achieve higher work efficiency; therefore, to improve the control of attrition. By creating different models, we are interested to see what caused employees from IBM to leave? To achieve this purpose, the executives and managers will be able to understand the current condition of the employees and take action to remedy controllable factors that can prevent attrition. Therefore, we intend to design an application which can help the Human Resource Department to further understand the structure of the employees who choose to leave and who stayed, and the attrition patterns regarding all features. From which can help not only to predict unwanted attrition, but to have proven action plans at your fingertips to help you reduce it, based on the organization's unique attributes.

Data Description

The data set is from IBM Community. There are 1471 entity instances in total with 30 attributes. Some of the information from the data set are recorded as numbers, we need to identify them as categorical variables and will be explained in details below.

Data Fields	Description	Datatype
Attrition	Turnover status of employee, stored as "Yes" or "No"	Binary
BusinessTravel	Whether the employee travel frequently or not	Categorical
DailyRate	The daily rate of employee	Numerical
Education	Education level of employee, 1 as 'Below College', 2 as 'College', 3 as 'Bachelor', 4 as 'Master', 5 as 'Doctor'	Categorical
EnvironmentSatisfication	The satisfication level of one employee in working environment, 1 as 'Low', 2 as 'Medium', 3 as 'High', 4 as 'Very High'	Categorical
JobInvolvement	Job involvement of employee, 1 as 'Low', 2 as 'Medium', 3 as 'High', 4 as 'Very High'	Categorical
JobLevel	Job level of employee	Numerical
JobSatisfaction	Job satisfaction of employee, 1 as 'Low', 2 as 'Medium', 3 as 'High', 4 as 'Very High'	Categorical
NumCompaniesWorked	Total number of companies the employee worked for	Numerical
OverTime	Work overtime or not	Binary
PercentSalaryHike	The rate of increase in income from last year to this year	Numerical
PerformanceRating	Rate of performance for employee, 1 as 'Low', 2 as 'Good', 3 as 'Excellent', 4 as 'Outstanding'	Categorical
RelationshipSatisfaction	Relationship satisfaction of employee, 1 as 'Low', 2 as 'Medium', 3 as 'High', 4 as 'Very High'	Categorical
StockOptionLevel	The number of stock employee hold	Numerical
TrainingTimesLastYear	Total training time for employee last year	Numerical
WorkLifeBalance	Does employee feel balance between work and life, 1 as 'Bad', 2 as 'Good', 3 as 'Better', 4 as 'Best'	Categorical

Critique of Existing Visualization

This radar chart above shows the performance and satisfaction level of employees in attrition or stay. It has revealed plenty of valuable information and is an excellent way to visualize data; however, only five features were evaluated consisting of ranges from 1 to 4. Though it might be easier to prepare, this chart might bring users wondering how other features compare these five. Therefore, a more precise way should be introduced to select elements constituting this chart. The best way to solve this problem is to use feature engineering, then to draw a new graph to visualize better the influence of different factors to attrition.

Visualization

Basic EDA Analytics will be used to preliminarily understand the pairwise relationship between two or three features and their effects on attrition. Based on the overall relations between all features, the Network Graph will be applied to understand the structural characteristics of the features as a whole and to select features which are fed for the further model. We also plan to use three methods including Decision Tree, Random Forest and XGBoost to build the attrition analytics model. Lastly, the User Portrait Analysis will be applied to understand the differences among groups more intuitively based on the most important features that are generated from the model result. We hope our users will have a clearer understanding of the distribution of employee attrition in different dimensions and get hints of how to improve their attrition management.

Dashboard Sketch

Methodology and Approach

In Feature engineering, we will generate variables from the previous ones and compose multiple features together, after which we will separate churn and not churn customer and separate categorical and numerical columns. The main point of this is selecting effective variables which would result in customer attrition. Apart from that, we may transform multi value variables to the dummy variable in the last step of this stage, we can obtain the variable summary like below:

Secondly, a correlation matrix of this model will be visualized to present the relationship among different variables and primarily understand the influence of each variables.

Proposed R Packages

Package Name	Description
shinydashboard	Enable the usage and design of shiny dashboard
shiny	Make the interactive web applications for data visualization
reshape	Give new shapes to an array without changing its data
plotly	Create interactive bar graphs and scatter plots
tidyverse	A set of packages to plot out various visualizations and EDA
readr	To read rectangular data
recharts	Create interactive radar chart
DT	Create data table
ggraph	An extension of ggplot2 to build plots layer by layer
corrgram	Create correlation matrix
ggthemes	To apply themes to Shiny applications
ggcorrplot	Visualize correlation matrix using ggplot2
plotrix	Create plot with two ordinates

Team Members

• Qiu Yang
• Shi Yawen
• Zhu Keyu

References

• Work Institute 2019 Retention Report final
• 7 Common (but Fixable) Causes of Employee Turnover
• Employee attrition has a “quick fix”