This is a HR analysis report in basis of the assignment from the Business Analytics course by ESSEC Business School. The analysis is implemented according to the phases, ASK, PREPARE, PROCESS, ANALYZE, SHARE, and ACT.
Data of Human Resource is anonymous data from a big consulting company that the number of employees leaves the firm. The Data is provided by ESSEC Business School for the purpose of the logistic analysis. Through out this analysis, we will check out the correlations and hierarchical clusters in the data set and understand what makes differences with two categories with visualizations.
Load following packages for the analysis work.
# Data manipulation packages
library(dplyr)
library(tidyverse)
library(scales)
library(skimr)
library(broom) #tidy the data table
# Visualization packages
library(ggplot2)
library(plotly)
library(kableExtra)
library(grid)
library(gridExtra)
library(GGally)
# Statistic packages
library(statsr)
library(corrplot)
library(PerformanceAnalytics)
library(caret)Import the data with csv file, “DATA_3.02_HR2.csv” and we see the first 6 rows of the data as follows.
df <- read_csv("DATA_3.02_HR2.csv")
head(df)# A tibble: 6 × 7
S LPE NP ANH TIC Newborn left
<dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
1 0.38 0.53 2 157 3 0 1
2 0.8 0.86 5 262 6 0 1
3 0.11 0.88 7 272 4 0 1
4 0.72 0.87 5 223 5 0 1
5 0.37 0.52 2 159 3 0 1
6 0.41 0.5 2 153 3 0 1Here, we load the the HR dataset as data frame and name as “df”. Accordingly, some variables are not easy to understand the meaning and we find all classes of values are numeric. We take a look at the statistic summary of the dataset.
| Name | df |
| Number of rows | 12000 |
| Number of columns | 8 |
| _______________________ | |
| Column type frequency: | |
| character | 1 |
| numeric | 7 |
| ________________________ | |
| Group variables | None |
Variable type: character
| skim_variable | n_missing | complete_rate | min | max | empty | n_unique | whitespace |
|---|---|---|---|---|---|---|---|
| ID | 0 | 1 | 1 | 5 | 0 | 12000 | 0 |
Variable type: numeric
| skim_variable | n_missing | complete_rate | mean | sd | p0 | p25 | p50 | p75 | p100 | hist |
|---|---|---|---|---|---|---|---|---|---|---|
| S | 0 | 1 | 0.63 | 0.24 | 0.09 | 0.48 | 0.66 | 0.82 | 1 | ▃▃▇▇▇ |
| LPE | 0 | 1 | 0.72 | 0.17 | 0.36 | 0.57 | 0.72 | 0.86 | 1 | ▂▇▆▇▇ |
| NP | 0 | 1 | 3.80 | 1.16 | 2.00 | 3.00 | 4.00 | 5.00 | 7 | ▇▆▃▁▁ |
| ANH | 0 | 1 | 200.44 | 48.74 | 96.00 | 157.00 | 199.50 | 243.00 | 310 | ▃▇▆▇▂ |
| TIC | 0 | 1 | 3.23 | 1.06 | 2.00 | 2.00 | 3.00 | 4.00 | 6 | ▅▇▃▂▁ |
| Newborn | 0 | 1 | 0.15 | 0.36 | 0.00 | 0.00 | 0.00 | 0.00 | 1 | ▇▁▁▁▂ |
| left | 0 | 1 | 0.17 | 0.37 | 0.00 | 0.00 | 0.00 | 0.00 | 1 | ▇▁▁▁▂ |
Accordingly, the dataset contains 12000 rows and seven columns with numeric class and ID column is character.There is not any missing data. “Newborn” and “left” are categorical variables because of binary input values.
Please refer to the table of definition for variable names as follows.
| Variable | Definition |
|---|---|
| S | Satisfaction rate of the employee about the company |
| LPE | Last project evaluation rate |
| NP | The number of completed projects within the last 12 months |
| ANH | Average number of hours worked per month |
| TIC | The time spent in the company |
| Newborn | Whether he or she had a baby within the last year |
| left | To indicate whether a employee left, (that is a 1), or stayed (that is a 0) |
| ID | Unique ID for employees |
Before we start analysis work, we take a look at the features of variables in the data.
In this step, we will apply for the pearson method as statistical approach to find out the strongest correlation with “left” variable.
According to the question, we need to find out the trends of left employees. As we see the correlation table, the following variables drive employees left.
Visualize relation between attrition and time in the company Let’s see the effect of one of the most important driver: TIC
According to the correlation analysis, satisfaction to the company is negatively impacted. The values of satisfaction rate has two decimals. To understand easily, we create categories of employee satisfaction ranking.
So far, we explored the features of variables, correlations, and visualizations between attrition and driving factors. From the data base, we found 16% of employees left from the company. “Time spent in the company” and “Satisfaction rate” are the main driving factors for the attrition. Then, who will leave a company at the high possibilities? To find out potential leaving employees are also the main purpose of this analysis. To answer to the question, we need to make a prediction model as well as visualizations of correlations of factors.
We use sample dataset of “DATA_4.02_HR3.csv” for testing model.
#Load sample data with ca.8% of data for testing.
#df_test <- read_csv("DATA_4.02_HR3.csv")
# convert the newborn to factor variables
#df_test=df_test %>%
# mutate(Newborn = as.factor(Newborn))
#head(df_test)list( summary_glm$coefficient, # Coefficient summary table about logistic regression to check p-value as significance.
round( 1 - ( summary_glm$deviance / summary_glm$null.deviance ), 4 ) ) #Formula for the pseudo R square value. How much the variances are explained by the model. [[1]]
Estimate Std. Error z value Pr(>|z|)
(Intercept) -1.107603517 0.1795189675 -6.169841 6.835862e-10
S -3.783653903 0.1358840924 -27.844716 1.248138e-170
LPE 0.453757223 0.2037831099 2.226667 2.596951e-02
NP -0.370137672 0.0297192759 -12.454465 1.322358e-35
ANH 0.003455542 0.0006998283 4.937700 7.904926e-07
TIC 0.611926710 0.0306594942 19.958800 1.256946e-88
Newborn1 -1.557233812 0.1312241103 -11.866979 1.756974e-32
[[2]]
[1] 0.2106In the data set, values of leaving company is the binary decision, such as “YES 1” or “NO 0”. In this case, the logistic regression model will be suitable for making a prediction about the probability of leaving company.
# prediction
df_train$prediction <- predict( model_glm, newdata = df_train, type = "response" )
df_test$prediction <- predict( model_glm, newdata = df_test , type = "response" )
# distribution of the prediction score grouped by known outcome
ggplot( df_train, aes( prediction, color = as.factor(left) ) ) +
geom_density( size = 1 ) +
ggtitle( "Training Set, Predicted Score" ) +
labs( color = "data" ) +
scale_color_discrete(labels = c("Negative", "Positive"))
As density chart shows, the predicted score is not so much accuracy yet. To make a classification, we apply for a cutoff value. Since the prediction of a logistic regression model is a probability, in order to use it as a classifier, we’ll have to choose a cutoff value as 0.3 tentatively.
# tidy from the broom package
coefficient <- tidy(model_glm)[ , c( "term", "estimate", "statistic" ) ]
# transfrom the coefficient to be in probability format
coefficient$estimate <- exp( coefficient$estimate )
coefficient# A tibble: 7 × 3
term estimate statistic
<chr> <dbl> <dbl>
1 (Intercept) 0.330 -6.17
2 S 0.0227 -27.8
3 LPE 1.57 2.23
4 NP 0.691 -12.5
5 ANH 1.00 4.94
6 TIC 1.84 20.0
7 Newborn1 0.211 -11.9 We have our logistic regression model as “model_glm”, we’ll load in the dataset again with outcome and use the model to predict the probability.
# use the model to predict a unknown outcome data "DATA_4.02_HR3.csv"
df_hr <- read_csv("DATA_4.02_HR3.csv" )Rows: 1000 Columns: 6
── Column specification ────────────────────────────────────────────────────────
Delimiter: ","
dbl (6): S, LPE, NP, ANH, TIC, Newborn
ℹ Use `spec()` to retrieve the full column specification for this data.
ℹ Specify the column types or set `show_col_types = FALSE` to quiet this message.df_hr$Newborn = as.factor(df_hr$Newborn)
# predict
df_hr$prediction <- predict( model_glm, newdata = df_hr, type = "response" )
list( head(df_hr), nrow(df_hr) )[[1]]
# A tibble: 6 × 7
S LPE NP ANH TIC Newborn prediction
<dbl> <dbl> <dbl> <dbl> <dbl> <fct> <dbl>
1 0.86 0.69 4 105 4 1 0.0137
2 0.52 0.98 4 209 2 0 0.103
3 0.84 0.6 5 207 2 0 0.0194
4 0.6 0.65 3 143 2 1 0.0174
5 0.85 0.57 3 227 2 0 0.0404
6 0.82 0.61 4 246 3 0 0.0613
[[2]]
[1] 1000we can use the cutoff value in the last section to determine who are the person to have an attention.
# cutoff (more than 30% of of probability of attrition)
cutoff = 0.3
df_hr <- df_hr[ df_hr$prediction >= cutoff, ]
list( head(df_hr), nrow(df_hr) )[[1]]
# A tibble: 6 × 7
S LPE NP ANH TIC Newborn prediction
<dbl> <dbl> <dbl> <dbl> <dbl> <fct> <dbl>
1 0.36 0.65 5 119 5 0 0.365
2 0.22 0.41 2 248 3 0 0.549
3 0.2 0.75 4 248 3 0 0.423
4 0.95 0.73 3 286 6 0 0.305
5 0.6 0.94 3 272 4 0 0.338
6 0.44 0.6 3 188 6 0 0.671
[[2]]
[1] 119
df_hr$LPECUT <- cut( df_hr$LPE, breaks = quantile(df_hr$LPE), include.lowest = TRUE )
median_lpe <- df_hr %>% group_by(LPECUT) %>%
summarise( prediction = median(prediction), count = n() )
ggplot( median_lpe, aes( LPECUT, prediction ) ) +
geom_point( aes( size = count ), color = "#51829B" ) +
theme( legend.position = "none" ) +
labs( title = "Last Project's Evaluation and Employee Attrition",
y = "Attrition Probability", x = "Last Project's Evaluation by Client" )+
theme_classic()
The employers needs to take an immediate action to the following employees.
df_meeting<- df_hr %>%
mutate(ID = row_number(),
prediction = percent(round(prediction, 4)),
Newborn = recode(Newborn, `0` = "No", `1` = "Yes")) %>%
arrange(desc(prediction)) %>%
filter(prediction > 60,
LPECUT != "[0.38,0.545]" ) %>% # Low evaluation workers are not so important.
select(ID, prediction,S, LPECUT, NP, ANH, TIC, Newborn)
df_meeting%>%
kable(col.names = c("ID", "Attrition Probability", "Satisfaction", "Evaluation Range", "Annual Project", "Average Monthly Working Hours", "Time Spent in Company", "Newborn"))%>%
kable_styling(bootstrap_options = "striped", "hoover")| ID | Attrition Probability | Satisfaction | Evaluation Range | Annual Project | Average Monthly Working Hours | Time Spent in Company | Newborn |
|---|---|---|---|---|---|---|---|
| 67 | 88.70% | 0.27 | (0.805,0.99] | 2 | 246 | 6 | No |
| 55 | 88.27% | 0.31 | (0.805,0.99] | 2 | 284 | 6 | No |
| 115 | 87.16% | 0.22 | (0.805,0.99] | 3 | 251 | 6 | No |
| 48 | 85.34% | 0.15 | (0.545,0.68] | 4 | 287 | 6 | No |
| 69 | 83.99% | 0.14 | (0.545,0.68] | 3 | 140 | 6 | No |
| 25 | 83.75% | 0.24 | (0.805,0.99] | 3 | 209 | 6 | No |
| 19 | 82.16% | 0.15 | (0.68,0.805] | 3 | 272 | 5 | No |
| 56 | 82.07% | 0.23 | (0.545,0.68] | 3 | 186 | 6 | No |
| 29 | 81.32% | 0.15 | (0.545,0.68] | 3 | 257 | 5 | No |
| 32 | 81.23% | 0.12 | (0.805,0.99] | 5 | 240 | 6 | No |
| 33 | 80.74% | 0.39 | (0.805,0.99] | 2 | 205 | 6 | No |
| 102 | 80.26% | 0.26 | (0.545,0.68] | 2 | 265 | 5 | No |
| 66 | 79.35% | 0.31 | (0.545,0.68] | 3 | 227 | 6 | No |
| 34 | 77.42% | 0.17 | (0.805,0.99] | 5 | 230 | 6 | No |
| 106 | 76.61% | 0.13 | (0.545,0.68] | 4 | 260 | 5 | No |
| 76 | 75.53% | 0.25 | (0.805,0.99] | 5 | 274 | 6 | No |
| 70 | 75.05% | 0.25 | (0.805,0.99] | 3 | 232 | 5 | No |
| 111 | 74.51% | 0.21 | (0.805,0.99] | 3 | 180 | 5 | No |
| 7 | 73.84% | 0.53 | (0.68,0.805] | 2 | 261 | 6 | No |
| 86 | 73.55% | 0.24 | (0.805,0.99] | 3 | 193 | 5 | No |
| 54 | 72.99% | 0.15 | (0.805,0.99] | 3 | 266 | 4 | No |
| 83 | 71.93% | 0.35 | (0.545,0.68] | 4 | 254 | 6 | No |
| 109 | 69.00% | 0.59 | (0.805,0.99] | 2 | 225 | 6 | No |
| 103 | 68.90% | 0.13 | (0.545,0.68] | 4 | 156 | 5 | No |
| 98 | 68.64% | 0.20 | (0.545,0.68] | 5 | 167 | 6 | No |
| 6 | 67.05% | 0.44 | (0.545,0.68] | 3 | 188 | 6 | No |
| 97 | 66.98% | 0.14 | (0.68,0.805] | 5 | 238 | 5 | No |
| 51 | 66.78% | 0.23 | (0.68,0.805] | 6 | 264 | 6 | No |
| 43 | 66.44% | 0.17 | (0.545,0.68] | 5 | 277 | 5 | No |
| 110 | 66.10% | 0.14 | (0.805,0.99] | 4 | 273 | 4 | No |
| 119 | 65.74% | 0.20 | (0.68,0.805] | 5 | 108 | 6 | No |
| 101 | 65.09% | 0.34 | (0.805,0.99] | 5 | 224 | 6 | No |
| 87 | 65.04% | 0.26 | (0.545,0.68] | 3 | 142 | 5 | No |
| 35 | 63.95% | 0.13 | (0.68,0.805] | 6 | 281 | 5 | No |
| 59 | 62.42% | 0.21 | (0.545,0.68] | 5 | 266 | 5 | No |
| 30 | 62.28% | 0.18 | (0.545,0.68] | 3 | 197 | 4 | No |
| 71 | 61.91% | 0.14 | (0.68,0.805] | 6 | 281 | 5 | No |
| 88 | 61.70% | 0.20 | (0.68,0.805] | 3 | 192 | 4 | No |
| 99 | 60.94% | 0.24 | (0.545,0.68] | 3 | 237 | 4 | No |
We hope this analysis will help to improve the attrition rate and to support the benefits of their business for long period.