Attrition Prediction

(Resignation Prediction)

Content

This is the capstone project for a data science immersive student. (DSI-1128)

Problem Statements

1. Employee attrition is a significant challenge for many organizations, leading to increased costs and decreased productivity.
2. The sample company has experienced a high employee turnover rate, and we want to develop an attrition prediction model that can identify employees at risk of leaving the company.
3. This project aims to develop a predictive model that can accurately identify employees who are likely to leave, allowing the company to take proactive measures to retain valuable employees and reduce the negative impact of attrition on the organization.
4. Goal of this project is to improve F1-score by 20% from the baseline model.

File

1. "README.md" (this file): explain the concept and contents of the project
2. Code folder: Contain all the coding, Data preparation, EDA, Baseline model, Feature selection model, and Hyper tuning model. And also contain pickle files of splitter train & test data which are already processed with Column Transfer.
3. Data folder: employee_data.csv: Original file from Kaggle, and the other files are files for transferring data between notebooks.
4. Image folder: Contains images for putting in the presentation.
5. Streamlit folder: File to use with Streamlit on both the local machine and online(streamlit.io).
6. "Capstone Project Attrition Prediction Resignation Prediction.pdf": Presentation slides.
7. LICENSE: Using MIT License

Data Dictionary

Feature	Type	Description
Age	int64	Age (unit range 18-60)
Attrition	object	Employee leaving the company (0=no, 1=yes)
BusinessTravel	object	1=No Travel, 2=Travel Frequently, 3=Travel Rarely
DailyRate	int64	Salary level (unit range 102-1499)
Department	object	Human Resources, Research & Development, Sales
DistanceFromHome	int64	Employee home distance from the office (unit range 1-29)
Education	int64	Level of education (1:Below College, 2:College, 3:Bachlor, 4:Master, 5:Doctor)(unit range 1-5)
EducationField	object	Human Resources, Technical Degree, Medical, Life Sciences, Marketing, Other
EmployeeCount	int64	Number of Employee for each observation(All are 1)
EmployeeNumber	int64	An Employee Number is a unique number that has been assigned to each current and former State employee and elected official in the Position and Personnel DataBase (PPDB).
EnvironmentSatisfaction	int64	Employee satisfaction with the company's environment (1:Low, 2:Medium, 3:High, 4:Very High) (unit range 1-4)
Gender	object	Male, Female
HourlyRate	int64	An hourly employee is paid a wage for each hour worked, unlike a salaried employee. (unit range 30-100)
JobInvolvement	int64	Employee involved in job (unit range 1-4)
JobLevel	int64	Job grades and classifications. (unit range 1-5)
JobRole	object	Role of employee: Human Resources, Research Director, Manufacturing Director, Healthcare Representative, Sales Executive, Sales Representative, Manager, Research Scientist, Laboratory Technician
JobSatisfaction	int64	Satisfaction with the job (unit range 1-4)
MaritalStatus	object	1=Divorced, 2=Married, 3=Single
MonthlyIncome	int64	Monthly salary
MonthlyRate	int64	Monthly rate
NumCompaniesWorked	int64	Number of companies worked at (unit range 1-9)
Over18	object	Y=yes, N=no (All are Y)
OverTime	object	1=yes, 0=no
PercentSalaryHike	int64	Percentage increase in salary
PerformanceRating	int64	Performance rating (1:Low, 2:Good, 3:Excellent, 4:Outstanding) (unit range 1-4)
RelationshipSatisfaction	int64	Relations satisfaction (1:Low, 2:Medium, 3:High, 4:Very High)(unit range 1-4)
StandardHours	int64	Standard hours (all are 80)
StockOptionLevel	int64	Employee stock options, also known as ESOs, are stock options in the company’s stock granted by an employer to certain employees.(unit range 0-3)
TotalWorkingYears	int64	Total years worked (unit range 0-40)
TrainingTimesLastYear	int64	Hours spent training last year (unit range 0-6)
WorkLifeBalance	int64	Time spent between work and outside (1:Bad, 2:Good, 3:Better, 4:Best) (unit range 1-4)
YearsAtCompany	int64	Total number of years at the company (unit range 0-40)
YearsInCurrentRole	int64	Years in current role (unit range 0-18)
YearsSinceLastPromotion	int64	Last promotion (unit range 0-15)
YearsWithCurrManager	int64	Years spent with the current manager (unit range 0-17)

Brief summary of Analysis

The improvements are shown in the following table

Score type	Baseline	Production	% of improvement
Recall	0.36	0.40	11%
Precision	0.59	0.88	49%
F1-score	0.45	0.55	22%

Confusion Matrix of the production model

Conclusion

1. The model met the expectations by improving the F1-Score by 22%. (Expected 20%)
2. Precision is 88% (if you guess 9 leave, 8 will be correct)

Recommendations

1. Any future improvements should prioritize high precision and acceptable recall.
2. To approach the employee, using this model carefully and with caution is important.