HR Analytics

HR Analytics and Employee performance IBM use of model classifiers to predict whether an employee is likely to quit or not.

Add library 💬

library(VIM)
library(plotrix)
library(dplyr)
library(tidyr)
library(tidyverse)
library(ggplot2)
library(lubridate)
library(corrplot)
library(shiny)
library(DT)
library(shinydashboard)
library(shinythemes)
library(ggthemes)
library(ggcorrplot)
library(cowplot)
library(caret)
library(rpart)
library(corrplot)
library(pROC)

Preview

Create decision tree and define importance of variables

data_ibm <- data
trainIndex <- createDataPartition(data_ibm$Attrition, p=0.80,list=FALSE, times=1)
train <- data_ibm[trainIndex,]
test <- data_ibm[-trainIndex,]

prop_train <- train %>% select(Attrition) %>%
  group_by(Attrition) %>% summarize(n=n()) %>%
  mutate(pct=round(prop.table(n), 2))

prop_test <- test %>% select(Attrition) %>% 
  group_by(Attrition) %>% summarize(n=n()) %>%
  mutate(pct=round(prop.table(n), 2))
rpart.tree <- rpart(Attrition ~ ., data=train)

var_imp <- data.frame(rpart.tree$variable.importance)
var_imp$features <- rownames(var_imp)
var_imp <- var_imp[, c(2, 1)]
var_imp$importance <- round(var_imp$rpart.tree.variable.importance, 2)
var_imp$rpart.tree.variable.importance <- NULL
colorCount <- length(unique(var_imp$features))

feature_importance <- var_imp %>%
  ggplot(aes(x=reorder(features, importance), y=importance, fill=features)) + geom_bar(stat='identity') + coord_flip() + 
  theme(legend.position = "none",plot.title = element_text(hjust = 0.5))+
  geom_label(aes(label=paste0(importance, "%")), colour = "white", fontface = "italic", hjust=0.6) + 
  labs(title="", x="Features", y="Importance")

Logistic Regression

Build the model in some simple steps as follows:

• Determine the independent variables
• Incorporating dependent variable "Attrition" in the model
• Find the variables most important in determining employee attrition
• Convert the model's data type from "character" to "formula"
• Incorporate tranining data into formulas and build models

Step 1: 35 Variables

# First, combine 30 variable dependencies to find meaningful variables
independentvariables1=c("Age", "BusinessTravel", "DailyRate", "Department", "DistanceFromHome", "Education", "EducationField", "EnvironmentSatisfaction", "Gender", "HourlyRate", "JobInvolvement", "JobLevel", "JobRole", "JobSatisfaction", "MaritalStatus", "MonthlyIncome", "MonthlyRate", "NumCompaniesWorked", "OverTime", "PercentSalaryHike", "PerformanceRating", "RelationshipSatisfaction", "StockOptionLevel", "TotalWorkingYears", "TrainingTimesLastYear", "WorkLifeBalance", "YearsAtCompany", "YearsInCurrentRole", "YearsSinceLastPromotion", "YearsWithCurrManager")
Model1=paste(independentvariables1,collapse="+")
Model_1=paste("Att~",Model1)
formula=as.formula(Model_1)

# Run Training Model1
Trainingmodel1=glm(formula=formula,data=TrainingData,family="binomial")
Trainingmodel1=step(object = Trainingmodel1,direction = "both")
summary(Trainingmodel1)

Step 2: 17 Variables

# After the first scoring iteration, choose the most significant value based on the P-value, which is 17 variables
# Combine 17 dependent variables to find meaningful variables
independentvariables2=c("BusinessTravel", "DistanceFromHome", "EnvironmentSatisfaction","JobInvolvement","JobRole","JobSatisfaction","MaritalStatus","NumCompaniesWorked","OverTime","RelationshipSatisfaction","TotalWorkingYears", "TrainingTimesLastYear","WorkLifeBalance","YearsAtCompany","YearsInCurrentRole", "YearsSinceLastPromotion","YearsWithCurrManager")
Model2=paste(independentvariables2,collapse="+")
Model_2=paste("Att~",Model2)
formula=as.formula(Model_2)

# Run Training Model2
Trainingmodel2=glm(formula=formula,data=TrainingData,family="binomial")
Trainingmodel2=step(object = Trainingmodel2,direction = "both")
summary(Trainingmodel2)

Step 3: 16 Variables

# After the second iteration of the calculation, remove the JobRole variable because the JobRoleSales Representative is the only JobRole that matters
# So we remove JobRole as a variant for our model.
# Combine 16 variable dependencies to find meaningful variables
independentvariables3=c("BusinessTravel", "DistanceFromHome", "EnvironmentSatisfaction","JobInvolvement","JobSatisfaction","MaritalStatus","NumCompaniesWorked","OverTime","RelationshipSatisfaction","TotalWorkingYears", "TrainingTimesLastYear","WorkLifeBalance","YearsAtCompany","YearsInCurrentRole", "YearsSinceLastPromotion","YearsWithCurrManager")
Model3=paste(independentvariables3,collapse="+")
Model_3=paste("Att~",Model3)
formula=as.formula(Model_3)

# Run Training Model3
Trainingmodel3=glm(formula=formula,data=TrainingData,family="binomial")
Trainingmodel3=step(object = Trainingmodel3,direction = "both")
summary(Trainingmodel3)

Step 4: 13 Variables

# After the third scoring iteration, WorkLifeBalance, YearsAtCompany, YearsWithCurrManager
# Combine 13 variable dependencies to find meaningful variables
independentvariables4=c("BusinessTravel", "DistanceFromHome", "EnvironmentSatisfaction","JobInvolvement","JobSatisfaction","MaritalStatus","NumCompaniesWorked","OverTime","RelationshipSatisfaction","TotalWorkingYears", "TrainingTimesLastYear","YearsInCurrentRole", "YearsSinceLastPromotion")
Model4=paste(independentvariables4,collapse="+")
Model_4=paste("Att~",Model4)
formula=as.formula(Model_4)

# Run Training Model4
Trainingmodel4=glm(formula=formula,data=TrainingData,family="binomial")
Trainingmodel4=step(object = Trainingmodel4,direction = "both")
summary(Trainingmodel4)

Graph the ROC for Training Model 1->4 and determine the AUC

troc1=roc(response=Trainingmodel1$y,predictor = Trainingmodel1$fitted.values,plot=T)
troc1$auc

troc2=roc(response=Trainingmodel2$y,predictor = Trainingmodel2$fitted.values,plot=T)
troc2$auc

troc3=roc(response=Trainingmodel3$y,predictor = Trainingmodel3$fitted.values,plot=T)
troc3$auc

troc4=roc(response=Trainingmodel4$y,predictor = Trainingmodel4$fitted.values,plot=T)
troc4$auc

🔗 Datasets

🐞 Thank You

• any kind of constructive suggestions are welcome , Please upvote this if you find it useful ✨