Introduction) Human Resource Analytics

Questions:

  • So why do our best and most experienced employees leave prematurely?
  • How can we normalize effort levels based on employee salary?
  • Would a better salary & work environment improve effort/satisfication levels?
  • Which sales industry performs better? What would happen if you were to introduce competition?
  • Can you anaylze the decision or preferences made by your empolyees?

Variables (That I make for clarfication):

  • satisfaction_level: Satisfaction level from (.09 - 1) probably filled in by employee.

  • last_evaluation: Last_evaluation (performance) level from (.36 - 1) probably filled in by the employee’s manager.

  • number_project: Number of projects that the employee has worked on.

  • average_monthly_hours: The average number of hours an employee worked in a month.

  • time_spend_company: Could this be the number of years an employee worked for this company?

  • Work_accident: Whether an employee has had an accident or not (0 = no, 1 = yes).

  • left: Whether an employee has left or not (0 = no, 1 = yes).

  • promoted_last_5years: Whether an employee has a promotion or not (0 = no, 1 = yes).

  • sales: The different department sectors of this company.

  • salary: Three-level salary ranging from (low, medium, high).

Descripitive Analytics) Exploring the Data

First I want to import and explore the Human Resource Dataset

HRanalytics <- read.csv("HR_analytics.csv", header = TRUE)

Always check for null values. In this example, we have specifically choosen a clean dataset (so there are no null values). However, not all dataset are clean and NA values can be a problem. When dealing with NA values you are left with 3 options keep it, remove it, or replace it. There are several techniques for handling null values, but we will cover them later down the line.

sum(is.na(HRanalytics))

Let’s start making comparisions:

summary(HRanalytics)

##  satisfaction_level last_evaluation  number_project  average_montly_hours
##  Min.   :0.0900     Min.   :0.3600   Min.   :2.000   Min.   : 96.0       
##  1st Qu.:0.4400     1st Qu.:0.5600   1st Qu.:3.000   1st Qu.:156.0       
##  Median :0.6400     Median :0.7200   Median :4.000   Median :200.0       
##  Mean   :0.6128     Mean   :0.7161   Mean   :3.803   Mean   :201.1       
##  3rd Qu.:0.8200     3rd Qu.:0.8700   3rd Qu.:5.000   3rd Qu.:245.0       
##  Max.   :1.0000     Max.   :1.0000   Max.   :7.000   Max.   :310.0       
##                                                                          
##  time_spend_company Work_accident         left       
##  Min.   : 2.000     Min.   :0.0000   Min.   :0.0000  
##  1st Qu.: 3.000     1st Qu.:0.0000   1st Qu.:0.0000  
##  Median : 3.000     Median :0.0000   Median :0.0000  
##  Mean   : 3.498     Mean   :0.1446   Mean   :0.2381  
##  3rd Qu.: 4.000     3rd Qu.:0.0000   3rd Qu.:0.0000  
##  Max.   :10.000     Max.   :1.0000   Max.   :1.0000  
##                                                      
##  promotion_last_5years         sales         salary    
##  Min.   :0.00000       sales      :4140   high  :1237  
##  1st Qu.:0.00000       technical  :2720   low   :7316  
##  Median :0.00000       support    :2229   medium:6446  
##  Mean   :0.02127       IT         :1227                
##  3rd Qu.:0.00000       product_mng: 902                
##  Max.   :1.00000       marketing  : 858                
##                        (Other)    :2923

## there are several ways of looking at grouped data [Fastest]
table(HRanalytics$sales)

## 
##  accounting          hr          IT  management   marketing product_mng 
##         767         739        1227         630         858         902 
##       RandD       sales     support   technical 
##         787        4140        2229        2720

## dpylr
HRanalytics %>% 
  count(left) 

## # A tibble: 2 × 2
##    left     n
##   <int> <int>
## 1     0 11428
## 2     1  3571

## plotting
plot(HRanalytics$salary)


## pairs [Slowest]
pairs(HRanalytics)

Predictive Analytics) Employee Retention

Training and Testing

##20, 80% split
train=HRanalytics[1:12000,]
test=HRanalytics[12001:nrow(HRanalytics),]

Decision Trees

fit = rpart(left~satisfaction_level+last_evaluation+number_project+average_montly_hours+time_spend_company+promotion_last_5years+sales+salary,data=train,method="class")
pred = predict(fit,test)

summary(fit)

## 0 = stayed, 1 = left
rpart.plot(fit, extra = 2, shadow.col ="gray")

Now we need to test its accuracy

printcp(fit)

## 
## Classification tree:
## rpart(formula = left ~ satisfaction_level + last_evaluation + 
##     number_project + average_montly_hours + time_spend_company + 
##     promotion_last_5years + sales + salary, data = train, method = "class")
## 
## Variables actually used in tree construction:
## [1] average_montly_hours last_evaluation      number_project      
## [4] satisfaction_level   time_spend_company  
## 
## Root node error: 2000/12000 = 0.16667
## 
## n= 12000 
## 
##        CP nsplit rel error xerror      xstd
## 1 0.16400      0    1.0000 1.0000 0.0204124
## 2 0.06375      3    0.4220 0.4220 0.0140057
## 3 0.05400      6    0.2270 0.2290 0.0104943
## 4 0.02000      7    0.1730 0.1750 0.0092167
## 5 0.01700      8    0.1530 0.1685 0.0090490
## 6 0.01050      9    0.1360 0.1395 0.0082540
## 7 0.01000     10    0.1255 0.1330 0.0080639

Random Forest

fit2 <- randomForest(left~satisfaction_level+last_evaluation+number_project+average_montly_hours+time_spend_company+promotion_last_5years+sales+salary,data=train)

## Warning in randomForest.default(m, y, ...): The response has five or fewer
## unique values. Are you sure you want to do regression?

print(fit2)

## 
## Call:
##  randomForest(formula = left ~ satisfaction_level + last_evaluation +      number_project + average_montly_hours + time_spend_company +      promotion_last_5years + sales + salary, data = train) 
##                Type of random forest: regression
##                      Number of trees: 500
## No. of variables tried at each split: 2
## 
##           Mean of squared residuals: 0.01532125
##                     % Var explained: 88.97

importance(fit2)

##                       IncNodePurity
## satisfaction_level        526.17718
## last_evaluation           194.39069
## number_project            295.31888
## average_montly_hours      246.58314
## time_spend_company        281.10927
## promotion_last_5years       1.69423
## sales                      23.11345
## salary                     13.82964

plot(fit2)


fit3 = rpart(left~satisfaction_level+number_project+average_montly_hours+time_spend_company,data=train,method="class")
pred = predict(fit,test)

## 0 = stayed, 1 = left
rpart.plot(fit3, extra = 2, shadow.col ="gray")

Overall, based on what we’ve learned from our tree models is that satisfaction_level plays the highest factor in predicting whether a employee leaves or not. It is then followed by number of projects given, time spent at the company, and average_monthly_hours. We know that employees would start to leave under these conditions. 4 or more projects with low satisfaction and 3 projects with extremely low satisfaction. The other condition is people working on less projects, but have incurred alot of overtime.

Data Visualization

GGPlot

ggplot(HRanalytics, aes(x = satisfaction_level)) +
geom_density(aes(colour=salary, fill=salary, alpha=.5))


ggplot(HRanalytics, aes(satisfaction_level, average_montly_hours, left))+ geom_density2d(aes(color = factor(left)))