员工离职预测
library(dplyr)
library(psych)
library(ggplot2)
library(randomForest)
str(train)
'data.frame': 1100 obs. of 31 variables: $ X...Age : int 37 54 34 39 28 24 29 36 33 34 ... $ Attrition : int 0 0 1 0 1 0 0 0 0 0 ... $ BusinessTravel : Factor w/ 3 levels "Non-Travel","Travel_Frequently",..: 3 2 2 3 2 3 3 3 3 3 ... $ Department : Factor w/ 3 levels "Human Resources",..: 2 2 2 2 2 3 2 3 2 2 ... $ DistanceFromHome : int 1 1 7 1 1 4 9 2 4 2 ... $ Education : int 4 4 3 1 3 1 5 2 4 4 ... $ EducationField : Factor w/ 6 levels "Human Resources",..: 2 2 2 2 4 4 5 4 4 6 ... $ EmployeeNumber : int 77 1245 147 1026 1111 1445 455 513 305 1383 ... $ EnvironmentSatisfaction : int 1 4 1 4 1 4 2 2 3 3 ... $ Gender : Factor w/ 2 levels "Female","Male": 2 1 2 1 2 1 2 2 1 1 ... $ JobInvolvement : int 2 3 1 2 2 3 2 2 2 3 ... $ JobLevel : int 2 3 2 4 1 2 1 3 1 2 ... $ JobRole : Factor w/ 9 levels "Healthcare Representative",..: 5 5 3 5 3 8 3 8 7 1 ... $ JobSatisfaction : int 3 3 3 4 2 3 4 3 2 4 ... $ MaritalStatus : Factor w/ 3 levels "Divorced","Married",..: 1 1 3 2 1 2 3 2 2 3 ... $ MonthlyIncome : int 5993 10502 6074 12742 2596 4162 3983 7596 2622 6687 ... $ NumCompaniesWorked : int 1 7 1 1 1 1 0 1 6 1 ... $ Over18 : Factor w/ 1 level "Y": 1 1 1 1 1 1 1 1 1 1 ... $ OverTime : Factor w/ 2 levels "No","Yes": 1 1 2 1 1 2 1 1 1 1 ... $ PercentSalaryHike : int 18 17 24 16 15 12 17 13 21 11 ... $ PerformanceRating : int 3 3 4 3 3 3 3 3 4 3 ... $ RelationshipSatisfaction: int 3 1 4 3 1 3 3 2 4 4 ... $ StandardHours : int 80 80 80 80 80 80 80 80 80 80 ... $ StockOptionLevel : int 1 1 0 1 2 2 0 2 0 0 ... $ TotalWorkingYears : int 7 33 9 21 1 5 4 10 7 14 ... $ TrainingTimesLastYear : int 2 2 3 3 2 3 2 2 3 2 ... $ WorkLifeBalance : int 4 1 3 3 3 3 3 3 3 4 ... $ YearsAtCompany : int 7 5 9 21 1 5 3 10 3 14 ... $ YearsInCurrentRole : int 5 4 7 6 0 4 2 9 2 11 ... $ YearsSinceLastPromotion : int 0 1 0 11 0 0 2 9 1 4 ... $ YearsWithCurrManager : int 7 4 6 8 0 3 2 0 1 11 ...
describe(train)
vars n mean sd median trimmed mad min max range skew kurtosis se X...Age 1 1100 37.00 9.04 36.0 36.51 8.90 18 60 42 0.44 -0.43 0.27 Attrition 2 1100 0.16 0.37 0.0 0.08 0.00 0 1 1 1.83 1.36 0.01 BusinessTravel* 3 1100 2.62 0.66 3.0 2.77 0.00 1 3 2 -1.47 0.81 0.02 Department* 4 1100 2.26 0.52 2.0 2.25 0.00 1 3 2 0.23 -0.41 0.02 DistanceFromHome 5 1100 9.43 8.20 7.0 8.36 7.41 1 29 28 0.91 -0.35 0.25 Education 6 1100 2.92 1.02 3.0 2.99 1.48 1 5 4 -0.30 -0.55 0.03 EducationField* 7 1100 3.22 1.32 3.0 3.06 1.48 1 6 5 0.58 -0.65 0.04 EmployeeNumber 8 1100 1028.16 598.92 1026.5 1027.04 782.81 1 2065 2064 0.02 -1.22 18.06 EnvironmentSatisfaction 9 1100 2.73 1.10 3.0 2.78 1.48 1 4 3 -0.33 -1.21 0.03 Gender* 10 1100 1.59 0.49 2.0 1.62 0.00 1 2 1 -0.38 -1.86 0.01 JobInvolvement 11 1100 2.73 0.71 3.0 2.74 0.00 1 4 3 -0.54 0.34 0.02 JobLevel 12 1100 2.05 1.11 2.0 1.89 1.48 1 5 4 1.04 0.40 0.03 JobRole* 13 1100 5.43 2.46 6.0 5.59 2.97 1 9 8 -0.34 -1.22 0.07 JobSatisfaction 14 1100 2.73 1.11 3.0 2.79 1.48 1 4 3 -0.33 -1.24 0.03 MaritalStatus* 15 1100 2.11 0.73 2.0 2.14 1.48 1 3 2 -0.18 -1.12 0.02 MonthlyIncome 16 1100 6483.62 4715.29 4857.0 5639.41 3166.09 1009 19999 18990 1.38 1.04 142.17 NumCompaniesWorked 17 1100 2.68 2.51 2.0 2.35 1.48 0 9 9 1.03 -0.02 0.08 Over18* 18 1100 1.00 0.00 1.0 1.00 0.00 1 1 0 NaN NaN 0.00 OverTime* 19 1100 1.28 0.45 1.0 1.22 0.00 1 2 1 0.99 -1.02 0.01 PercentSalaryHike 20 1100 15.24 3.63 14.0 14.85 2.97 11 25 14 0.79 -0.35 0.11 PerformanceRating 21 1100 3.15 0.36 3.0 3.07 0.00 3 4 1 1.93 1.72 0.01 RelationshipSatisfaction 22 1100 2.70 1.10 3.0 2.75 1.48 1 4 3 -0.29 -1.23 0.03 StandardHours 23 1100 80.00 0.00 80.0 80.00 0.00 80 80 0 NaN NaN 0.00 StockOptionLevel 24 1100 0.79 0.84 1.0 0.67 1.48 0 3 3 0.95 0.34 0.03 TotalWorkingYears 25 1100 11.22 7.83 10.0 10.27 5.93 0 40 40 1.15 0.99 0.24 TrainingTimesLastYear 26 1100 2.81 1.29 3.0 2.74 1.48 0 6 6 0.50 0.49 0.04 WorkLifeBalance 27 1100 2.75 0.70 3.0 2.76 0.00 1 4 3 -0.60 0.47 0.02 YearsAtCompany 28 1100 7.01 6.22 5.0 5.94 4.45 0 37 37 1.81 4.01 0.19 YearsInCurrentRole 29 1100 4.21 3.62 3.0 3.83 4.45 0 18 18 0.95 0.61 0.11 YearsSinceLastPromotion 30 1100 2.23 3.31 1.0 1.49 1.48 0 15 15 1.94 3.30 0.10 YearsWithCurrManager 31 1100 4.12 3.60 3.0 3.76 4.45 0 17 17 0.86 0.26 0.11
#删除 常量
name<-names(train)
train<-train[name!="Over18" & name!="StandardHours" & name!="EmployeeNumber"]
#重编码
train$Gender<-as.integer(train$Gender)-1
train$OverTime<-as.integer(train$OverTime)-1
#Age 和 Attrition
ggplot(train, aes(X...Age, fill = factor(Attrition))) +
geom_histogram(bins=30) +
facet_grid(.~Gender)+
labs(fill="Attrition")+ xlab("Age")+ylab("Total Count")
#小结：
train$X...Age[train$X...Age>=18 & train$X...Age <25]<-1 train$X...Age[train$X...Age>=25 & train$X...Age <35]<-2 train$X...Age[train$X...Age>=35 & train$X...Age <45]<-3 train$X...Age[train$X...Age>=45 & train$X...Age <55]<-4 train$X...Age[train$X...Age>=55 ]<-5
#Department 和 JobLevel
ggplot(train, aes(x = JobLevel, fill = as.factor(Attrition))) + geom_bar() + facet_wrap(~ Department)+
xlab("Job Level")+
ylab("Total Count")+
labs(fill = "Attrition")
train$Department<-as.character(train$Department) train$Department[train$Department=="Human Resources"]<-"1" train$Department[train$Department=="Sales"]<-"2" train$Department[train$Department=="Research & Development"]<-"3" train$Department<-as.integer(train$Department)
#小结：不同部门相同级别之间存在明显差异，研发部门1，2级别和销售部1，2，3级别流动性较大。
#Department 和 BusinessTravel
ggplot(train, aes(x = BusinessTravel, fill = as.factor(Attrition))) + geom_bar() + facet_wrap(~ Department)+ xlab("BusinessTravel")+ ylab("Total Count")+ labs(fill = "Attrition")
train$BusinessTravel<-as.character(train$BusinessTravel) train$BusinessTravel[train$BusinessTravel=="Non-Travel"]<-"1" train$BusinessTravel[train$BusinessTravel=="Travel_Frequently"]<-"2" train$BusinessTravel[train$BusinessTravel=="Travel_Rarely"]<-"3" train$BusinessTravel<-as.integer(train$BusinessTravel)
#小结：是否经常出差，并不是影响离职的关键因素，但偶然出差的员工离职率最高。研发部、销售部、人力资源部依次下降。
#EducationField 和 Attrition
ggplot(train,aes(EducationField,fill=as.factor(Attrition)))+ geom_bar(stat="count",position="dodge")+ xlab("EducationField")+ ylab("Total Count")+ labs(fill="Attrition")
#小结：专业领域和离职之间无明显关系
#MaritalStatus 和 Attrition
ggplot(train,aes(MaritalStatus,fill=as.factor(Attrition)))+
geom_bar(stat="count",position="dodge")+
xlab("MaritalStatus")+
ylab("Total Count")+
labs(fill="Attrition")
train$MaritalStatus<-as.character(train$MaritalStatus) train$MaritalStatus[train$MaritalStatus=="Divorced"]<-1 train$MaritalStatus[train$MaritalStatus=="Married"]<-2 train$MaritalStatus[train$MaritalStatus=="Single"]<-3 train$MaritalStatus<-as.integer(train$MaritalStatus)
#小结：婚姻情况和离职有一点关系
#EnvironmentSatisfaction 和 Attrition
ggplot(train, aes(x = EnvironmentSatisfaction, fill = as.factor(Attrition))) + geom_bar() + facet_wrap(~ JobLevel)+ xlab("JobLevel")+ ylab("Total Count")+ labs(fill = "Attrition")
#小结：满意度和离职之间无明显关系
#MonthlyIncome 和 Attrition
ggplot(train,aes(MonthlyIncome, fill = factor(Attrition))) + geom_density(alpha = 0.8)+ labs(fill="Attrition")
#小结：低收入者明显在职意向不稳定
train$MonthlyIncome[train$MonthlyIncome<=3000]<-1 train$MonthlyIncome[train$MonthlyIncome>3000 & train$MonthlyIncome<=6000]<-2 train$MonthlyIncome[train$MonthlyIncome>6000 & train$MonthlyIncome<=9000]<-3 train$MonthlyIncome[train$MonthlyIncome>9000 & train$MonthlyIncome<=12000]<-4 train$MonthlyIncome[train$MonthlyIncome>12000 & train$MonthlyIncome<=17000]<-5 train$MonthlyIncome[train$MonthlyIncome>17000]<-6
#关联关系
corrgram(train[,-c(7,12)],lower.panel=panel.pie,upper.panel=NULL)
#抽样
set.seed(1) ind<-sample(2,nrow(train),replace=TRUE,prob=c(0.7,0.3)) train.df<-train[ind==1,] test.df<-train[ind==2,]
#随机森林
rf<-randomForest(factor(Attrition)~.,data=train.df) varImpPlot(rf)
#准确率
prediction <- predict(rf,newdata=test.df,type="response") misClasificError <- mean(prediction != test.df$Attrition) print(paste('Accuracy',1-misClasificError)) [1] "Accuracy 0.876506024096386"
#逻辑回归
gf<-glm(Attrition~.,data=train.df,family = binomial(link=logit))
summary(gf)
Call:
glm(formula = Attrition ~ ., family = binomial(link = logit),
data = train.df)
Deviance Residuals:
Min 1Q Median 3Q Max
-1.6113 -0.5048 -0.2459 -0.0860 3.4737
Coefficients:
Estimate Std. Error z value Pr(>|z|)
(Intercept) -1.31753 3.58064 -0.368 0.712903
X...Age -0.36136 0.17640 -2.049 0.040508 *
BusinessTravel 0.05908 0.20110 0.294 0.768928
Department 0.39070 0.93431 0.418 0.675820
DistanceFromHome 0.05274 0.01486 3.550 0.000386 ***
Education -0.17039 0.12279 -1.388 0.165239
EducationFieldLife Sciences -0.43924 1.20639 -0.364 0.715785
EducationFieldMarketing 0.14995 1.25574 0.119 0.904948
EducationFieldMedical -0.55928 1.20602 -0.464 0.642835
EducationFieldOther 0.07420 1.32247 0.056 0.955256
EducationFieldTechnical Degree 0.62904 1.22665 0.513 0.608084
EnvironmentSatisfaction -0.50299 0.11646 -4.319 1.57e-05 ***
Gender 0.49495 0.26618 1.859 0.062965 .
JobInvolvement -0.67266 0.17777 -3.784 0.000154 ***
JobLevel -0.18383 0.39279 -0.468 0.639777
JobRoleHuman Resources 2.92472 2.10883 1.387 0.165474
JobRoleLaboratory Technician 2.11121 0.82806 2.550 0.010785 *
JobRoleManager 2.09557 1.12796 1.858 0.063193 .
JobRoleManufacturing Director 1.22695 0.84649 1.449 0.147211
JobRoleResearch Director 1.49258 1.17894 1.266 0.205501
JobRoleResearch Scientist 1.44543 0.82801 1.746 0.080868 .
JobRoleSales Executive 2.17131 1.20040 1.809 0.070479 .
JobRoleSales Representative 3.29933 1.28712 2.563 0.010367 *
JobSatisfaction -0.63089 0.11767 -5.361 8.26e-08 ***
MaritalStatus 0.94530 0.25146 3.759 0.000170 ***
MonthlyIncome -0.03459 0.24924 -0.139 0.889628
NumCompaniesWorked 0.13934 0.05418 2.572 0.010119 *
OverTime 2.18546 0.27520 7.941 2.00e-15 ***
PercentSalaryHike -0.05939 0.05572 -1.066 0.286492
PerformanceRating 0.86885 0.55923 1.554 0.120266
RelationshipSatisfaction -0.33278 0.11625 -2.863 0.004201 **
StockOptionLevel 0.01585 0.21361 0.074 0.940835
TotalWorkingYears -0.04047 0.04220 -0.959 0.337593
TrainingTimesLastYear -0.15291 0.10058 -1.520 0.128425
WorkLifeBalance -0.21648 0.16944 -1.278 0.201398
YearsAtCompany 0.07885 0.05340 1.477 0.139745
YearsInCurrentRole -0.13861 0.06449 -2.149 0.031612 *
YearsSinceLastPromotion 0.14022 0.05867 2.390 0.016857 *
YearsWithCurrManager -0.11790 0.06483 -1.819 0.068956 .
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
(Dispersion parameter for binomial family taken to be 1)
Null deviance: 717.22 on 767 degrees of freedom
Residual deviance: 457.00 on 729 degrees of freedom
AIC: 535
Number of Fisher Scoring iterations: 6
#准确率
prediction <- predict(gf,newdata=test.df,type="response") prediction <- ifelse(prediction > 0.5,1,0) misClasificError <- mean(prediction != test.df$Attrition) print(paste('Accuracy',1-misClasificError)) [1] "Accuracy 0.858433734939759"