EMPLOYEE RETENTION

Objective:

HR Team works on better anticipating unplanned leave or retirement in our Parisian workforce

Workflow

1. Data Cleaning

2. Exploratory Analysis

3. Modelling

4. Takeaways

1338 rows
35 columns
26 Numerical columns
9 Categorical columns
No NaN values

After cleaning, our dataset shows 214 resignations (19%) over these past 2 years

y = df.Attrition (0, 1) = Resigned or not

3 retention strategies

Data Cleaning

```
Drop repetitive columns
```

Winsorize 6 columns by visualising box plots

Create dummies for Gender, MaritalStatus & Department, drop=True

Numerize Attrition & OverTime ({'Yes':1,'No':0}) and BusinessTravel ({'Travel_Rarely':1,'Travel_Frequently':2, 'Non-Travel':0})

Create bins for Age ( 30>40>50>60 ), drop=True

==> total of 27 numerical columns

YearsAtCompany Violinplot

Attrition 1 = Employee Left

Insight:
< 5 years of experience at Sanofi, more chance to leave

YearsWithCurrManager & YearsInCurrentRole

Violinplots

Attrition 1 = Employee Resigned

Insights:
After 2 YearsWithCurrManager, after 2 YearsInCurrentRole... more people staying

JobLevel & StockOptionLevel

Violinplots
Attrition 1 = Employee Left

JobLevel 1 = Entry-Level position

Insights:
The higher the JobLevel and Benefit Packages ... the less chance of leaving

BusinessTravel & OverTime

Violinplots
Attrition 1 = Employee Left

1 = Casual Businesstrip or overtime

Insights:
Travelling for work and doing overtime affects retention ... less resignations for 0

WorkLifeBalance & Age

Violinplots

Attrition 1 = Employee Left
WorkLifeBalance 4 = Employee highly values

Insights:
< 30 years old or Work-life balance highly valued... more risk to see the employee leave

Other Insights

No impact

PercentSalaryHike; DistanceFromHome; HourlyRate; NumCompaniesWorked; YearsSinceLastPromotion;

EnvironmentSatisfaction; 
JobSatisfaction;

People say what you wanna hear

EnvironmentSatisfaction & JobSatisfaction,
probably explained by a response bias
(internal survey)

Let's check these assumptions with our models

VIF Test performed

Remaining columns saved in X variables

Correlation Matrix for X

Input Data Selection

Input Data

MODELS

Priorities: 
1. Maximise AUC, the model's precision
2. Minimise False Negative rate (= predicted no resignation but the employee left)

# Model 1: Logistic Regression
# Model 2: Logistic Regression w/ normalizing HourlyRate
# Model 3 & 4: KNearest Neighbors
# Model 5: Naive Bayes Model
# Model 6 & 7: Decision Tree
# Model 8 & 9: Random Forest
# Model 10: Support Vector Machine
# Model 11: Nu SVC

For every model

lst=[]
for i in range(1,13):
    FP = eval(f'conf{i}')[0][1]
    FN = eval(f'conf{i}')[1][0]
    TP = eval(f'conf{i}')[1][1]
    TN = eval(f'conf{i}')[0][0]
    
    FNR = FN/(TP+FN)*100
    FPR = FP/(FP+TN)*100
    ACC = (TP+TN)/(TP+FP+FN+TN)*100
    AUC = eval(f'model{i}_roc')
    
    lst.append([i,FNR,FPR,ACC,AUC])
    
results = pd.DataFrame(lst, columns=['Model','False_negative_rate',
		'False_Positive_rate','Overall_Accuracy','Area_Under_Curve'])

False Negative Rate         False Positive Rate

Area Under the Curve        Overall Accuracy

Model Evaluation

Priorities: 
1. Maximise AUC, the model's precision
2. Minimise False Negative rate (= predicted no resignation but the employee left)

Logistic Regression with PCA

Priorities: 
1. Maximise AUC, the model's precision
2. Minimise False Negative rate (= predicted no resignation but the employee left)

# Model 1: Logistic Regression
# Model 12: Logistic Regression with X_train_PCA & X_test_PCA



AUC = 0.54 & FNR = 54.6

We keep our X

FINAL MODEL: Logistic Regression without PCA

AUC = 0.71 & FNR = 28

Expression:

0.64*BusinessTravel + 0.033*DistanceFromHome - 0.046*Education - 0.39*EnvironmentSatisfaction + 0.003*HourlyRate - 0.4*JobLevel - 0.39*JobSatisfaction + 0.14*NumCompaniesWorked' + 1.4*OverTime - 0.69*StockOptionLevel - 0.07*TrainingTimesLastYear - 0.01*YearsInCurrentRole + 0.02*YearsSinceLastPromotion - - 0.11*YearsWithCurrManager + 0.34*Gender_Male + 0.23*MaritalStatus_Married + 0.44*MaritalStatus_Single + 0.69*Department_Sales - 0.29*Age_bins

Employee Retention Strategy

Key take aways: 

1. Major focus on Sales department (high turnover) and entry-level position

2. Minimise business trips and anticipate high-demand cycles to  foster a work environment pro work-life balance

3. Minimise overtime as much as possible and reward with stock options or training sessions

deck

By Alexis Lacabane

deck

More from Alexis Lacabane