Skip to content
/ Prediction_of_Diabetes_using_R Public

Identified the optimal order of preprocessing for Pima Diabetes Dataset, how feature selection and upsampling affects the evaluation metrics

0 stars 0 forks Branches Tags Activity
Star
Notifications You must be signed in to change notification settings

TejaswiniKatale/Prediction_of_Diabetes_using_R

BranchesTags

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

3 Commits
README.md
README.md
 
 

Repository files navigation

Prediction_of_Diabetes_using_R

###1)Without Feature selection and then Upsampling### library(caTools) f<-read.csv("D:\Excel\DiabetesDiabetes_1.csv _1.csv") f<-f[,c(-1,-2)]

#data cleaning---- #remove duplicate values
f<-unique(f)

#Fixing and standardizing typos #1. Gender for(i in 1:length(f$Gender)){ if (f$Gender[i]=='M'){ f$Gender[i]=0; } else{ f$Gender[i]=1; } } #2. CLASS for(i in 1:length(f$CLASS)){ if (f$CLASS[i]=='N'){ f$CLASS[i]=1; } else if(f$CLASS[i]=='P'){ f$CLASS[i]=2; } else{ f$CLASS[i]=3; } } f<-transform(f,Gender=as.integer(Gender),CLASS=as.integer(CLASS));

#dividing into training and testing data---- set.seed(123) split<-sample.split(f,SplitRatio = 0.7) tr_data<-subset(f,split==TRUE) ts_data<-subset(f,split==FALSE)

#DECISION TREE---- library(rpart) model2 <- rpart(CLASS ~ . , data=tr_data,method="class")

  #plot(model2, uniform=TRUE, main="Classification Tree for Diabetes")
  #text(model2, use.n=TRUE, all=TRUE, cex=.8)
  
  treePred <- predict(model2, ts_data, type = 'class')
  tb1 = table(treePred, ts_data$CLASS)
  cat("\n")
  cat("\nDecison tree :")
  #Accuracy
  accuracy_Test <- sum(diag(tb1)) / sum(tb1)
  cat("\n    Accuracy :",accuracy_Test)
  acc_vector = c(accuracy_Test)

#Precision pre91<-(tb1[1,1]/sum(tb1[,1])) cat("\n Precision of Class N :",pre91) pre92<-(tb1[2,2]/sum(tb1[,2])) cat("\n Precision of Class P :",pre92) pre93<-(tb1[3,3]/sum(tb1[,3])) cat("\n Precision of Class Y :",pre93)

  prec_n = c(pre91)
  prec_p = c(pre92)
  prec_y = c(pre93)

#Recall rec91<-(tb1[1,1]/sum(tb1[1,])) cat("\n Recall of Class N :",rec91) rec92<-(tb1[2,2]/sum(tb1[2,])) cat("\n Recall of Class P :",rec92) rec93<-(tb1[3,3]/sum(tb1[3,])) cat("\n Recall of Class Y :",rec93,"\n")

  rec_n = c(rec91)
  rec_p = c(rec92)
  rec_y = c(rec93)

#KNN----

  tr1_data = tr_data
  ts1_data = ts_data

#Normalize tr1_data[,2:11]<-scale(tr1_data[,2:11]) ts1_data[,2:11]<-scale(ts1_data[,2:11])

#Apply knn library(class) y_pred<-knn(train = tr1_data[-12], test = ts1_data[-12], cl = tr1_data[,12], k = 3)

#Confusion Matrix cf1<-table(actual=ts1_data$CLASS,predicted = y_pred)

  cat("\nKNN :")
  #accuracy
  acc<-(cf1[[1,1]]+cf1[[2,2]]+cf1[[3,3]])/sum(cf1)
  cat("\n    Accuracy :",acc)
  acc_vector = c(acc_vector,acc)
  
  #Precision
  pre1<-(cf1[1,1]/sum(cf1[,1]))
  cat("\n    Precision of Class N :",pre1)
  pre2<-(cf1[2,2]/sum(cf1[,2]))
  cat("\n    Precision of Class P :",pre2)
  pre3<-(cf1[3,3]/sum(cf1[,3]))
  cat("\n    Precision of Class Y :",pre3)

  prec_n = c(prec_n,pre1)
  prec_p = c(prec_p,pre2)
  prec_y = c(prec_y,pre3)
  
  #Recall
  rec1<-(cf1[1,1]/sum(cf1[1,]))
  cat("\n    Recall of Class N :",rec1)
  rec2<-(cf1[2,2]/sum(cf1[2,]))
  cat("\n    Recall of Class P :",rec2)
  rec3<-(cf1[3,3]/sum(cf1[3,]))
  cat("\n    Recall of Class Y :",rec3,"\n")

  rec_n = c(rec_n,rec1)
  rec_p = c(rec_p,rec2)
  rec_y = c(rec_y,rec3)

#RANDOM fOREST---- library("randomForest")

  ts2_data<-ts_data
  tr2_data<-tr_data
  set.seed(345)

#Applying random forest rfm= randomForest(CLASS~.,data = tr2_data,ntree=10) y_pred1= predict(rfm,ts2_data)

  y_pred1[y_pred1<1.5]=1
  y_pred1[1.5<=y_pred1 & y_pred1<2.5]=2
  y_pred1[2.5<=y_pred1]=3

  ts2_data$y_pred1 = y_pred1

#Confusion Matrix cf2<-table(actual=ts2_data$CLASS,predicted = y_pred1)

  cat("\nRANDOM FOREST:")

#accuracy acc2<-(cf2[[1,1]]+cf2[[2,2]]+cf2[[3,3]])/sum(cf2) cat("\n Accuracy:",acc2) acc_vector = c(acc_vector,acc2)

#Precision pre12<-(cf2[1,1]/sum(cf2[,1])) cat("\n Precision of Class N :",pre12) pre22<-(cf2[2,2]/sum(cf2[,2])) cat("\n Precision of Class P :",pre22) pre33<-(cf2[3,3]/sum(cf2[,3])) cat("\n Precision of Class Y :",pre33)

  prec_n = c(prec_n,pre12)
  prec_p = c(prec_p,pre22)
  prec_y = c(prec_y,pre33)

  #Recall
  rec12<-(cf2[1,1]/sum(cf2[1,]))
  cat("\n    Recall of Class N :",rec12)
  rec22<-(cf2[2,2]/sum(cf2[2,]))
  cat("\n    Recall of Class P : ",rec22)
  rec33<-(cf2[3,3]/sum(cf2[3,]))
  cat("\n    Recall of Class Y :",rec33)
  cat("\n")      
  cat("\n")
  cat("\n")
  
  rec_n = c(rec_n,rec12)
  rec_p = c(rec_p,rec22)
  rec_y = c(rec_y,rec33)

Visualization---

  plot(acc_vector,type="o",col="red",xlab="Algorithms",ylab="Accuracy",xaxt='n',lwd=5,cex.lab=1.5)
  axis(1,at=c(1,2,3),labels=c("DT","KNN","RF"),cex.axis=2)

  plot(prec_n,type="o",col="red",xlab="Algorithms",ylab="Precision",ylim=c(0,1),xaxt='n',lwd=5,cex.lab=1.5)
  lines(prec_p, type="o",col="blue",lwd=5)
  lines(prec_y, type="o",col="green",lwd=5)
  legend(1,0.55,legend = c("Class N","Class P","Class Y"),fill = c("red","blue","green"),cex=1)
  axis(1,at=c(1,2,3),labels=c("DT","KNN","RF"),cex.axis=2)
  
  plot(rec_n,type="o",col="red",xlab="Algorithms",ylab="Recall",ylim=c(0,1),xaxt='n',lwd=5,cex.lab=1.5)
  lines(rec_p, type="o",col="blue",lwd=5)
  lines(rec_y, type="o",col="green",lwd=5)
  legend(1,0.55,legend = c("Class N","Class P","Class Y"),fill = c("red","blue","green"),cex=1)
  axis(1,at=c(1,2,3),labels=c("DT","KNN","RF"),cex.axis=2)

###2)With Upsampling then Feature Selection ### library(caTools) f<-read.csv("D:\Excel\Diabetes_1.csv") f<-f[,c(-1,-2)]

#data cleaning---- #remove duplicate values
f<-unique(f)

#Fixing and standardizing typos #1. Gender for(i in 1:length(f$Gender)){ if (f$Gender[i]=='M'){ f$Gender[i]=0; } else{ f$Gender[i]=1; } } #2. CLASS for(i in 1:length(f$CLASS)){ if (f$CLASS[i]=='N'){ f$CLASS[i]=1; } else if(f$CLASS[i]=='P'){ f$CLASS[i]=2; } else{ f$CLASS[i]=3; } } f<-transform(f,Gender=as.integer(Gender),CLASS=as.integer(CLASS));

#upsampling---- N<-which(f$CLASS==1) P<-which(f$CLASS==2) Y<-which(f$CLASS==3) length(N) length(P) length(Y)

  set.seed(121)
  Up.sampling1<-sample(N,length(Y),replace = TRUE)
  Up.sampling2<-sample(P,length(Y),replace = TRUE)
  f<-f[c(Up.sampling1,Up.sampling2,Y),]

#feature selection---- library(Boruta) set.seed(1234) imp_features <- Boruta(CLASS~.,data=f,doTrace=2) print(imp_features)

#Removing unimportant features # No unimportant features found

#dividing into training and testing data---- set.seed(123) split<-sample.split(f,SplitRatio = 0.7) tr_data<-subset(f,split==TRUE) ts_data<-subset(f,split==FALSE)

#DECISION TREE---- library(rpart) model2 <- rpart(CLASS ~ . , data=tr_data,method="class")

  #plot(model2, uniform=TRUE, main="Classification Tree for Diabetes")
  #text(model2, use.n=TRUE, all=TRUE, cex=.8)
  
  treePred <- predict(model2, ts_data, type = 'class')
  tb1 = table(treePred, ts_data$CLASS)
  cat("\n")
  cat("\nDecison tree :")

#Accuracy accuracy_Test <- sum(diag(tb1)) / sum(tb1) cat("\n Accuracy :",accuracy_Test) acc_vector = c(accuracy_Test)

#Precision pre91<-(tb1[1,1]/sum(tb1[,1])) cat("\n Precision of Class N :",pre91) pre92<-(tb1[2,2]/sum(tb1[,2])) cat("\n Precision of Class P :",pre92) pre93<-(tb1[3,3]/sum(tb1[,3])) cat("\n Precision of Class Y :",pre93)

  prec_n = c(pre91)
  prec_p = c(pre92)
  prec_y = c(pre93)

#Recall rec91<-(tb1[1,1]/sum(tb1[1,])) cat("\n Recall of Class N :",rec91) rec92<-(tb1[2,2]/sum(tb1[2,])) cat("\n Recall of Class P :",rec92) rec93<-(tb1[3,3]/sum(tb1[3,])) cat("\n Recall of Class Y :",rec93,"\n")

  rec_n = c(rec91)
  rec_p = c(rec92)
  rec_y = c(rec93)

#KNN----

  tr1_data = tr_data
  ts1_data = ts_data

#Normalize tr1_data[,2:11]<-scale(tr1_data[,2:11]) ts1_data[,2:11]<-scale(ts1_data[,2:11])

#Apply knn library(class) y_pred<-knn(train = tr1_data[-12], test = ts1_data[-12], cl = tr1_data[,12], k = 3)

#Confusion Matrix cf1<-table(actual=ts1_data$CLASS,predicted = y_pred)

  cat("\nKNN :")
  #accuracy
  acc<-(cf1[[1,1]]+cf1[[2,2]]+cf1[[3,3]])/sum(cf1)
  cat("\n    Accuracy :",acc)
  acc_vector = c(acc_vector,acc)
  
  #Precision
  pre1<-(cf1[1,1]/sum(cf1[,1]))
  cat("\n    Precision of Class N :",pre1)
  pre2<-(cf1[2,2]/sum(cf1[,2]))
  cat("\n    Precision of Class P :",pre2)
  pre3<-(cf1[3,3]/sum(cf1[,3]))
  cat("\n    Precision of Class Y :",pre3)
  
  prec_n = c(prec_n,pre1)
  prec_p = c(prec_p,pre2)
  prec_y = c(prec_y,pre3)
  
  #Recall
  rec1<-(cf1[1,1]/sum(cf1[1,]))
  cat("\n    Recall of Class N :",rec1)
  rec2<-(cf1[2,2]/sum(cf1[2,]))
  cat("\n    Recall of Class P :",rec2)
  rec3<-(cf1[3,3]/sum(cf1[3,]))
  cat("\n    Recall of Class Y :",rec3,"\n")
  
  rec_n = c(rec_n,rec1)
  rec_p = c(rec_p,rec2)
  rec_y = c(rec_y,rec3)

#RANDOM fOREST---- library("randomForest")

  ts2_data<-ts_data
  tr2_data<-tr_data
  set.seed(345)

#Applying random forest rfm= randomForest(CLASS~.,data = tr2_data,ntree=10) y_pred1= predict(rfm,ts2_data)

  y_pred1[y_pred1<1.5]=1
  y_pred1[1.5<=y_pred1 & y_pred1<2.5]=2
  y_pred1[2.5<=y_pred1]=3

  ts2_data$y_pred1 = y_pred1

#Confusion Matrix cf2<-table(actual=ts2_data$CLASS,predicted = y_pred1)

  cat("\nRANDOM FOREST:")

#accuracy acc2<-(cf2[[1,1]]+cf2[[2,2]]+cf2[[3,3]])/sum(cf2) cat("\n Accuracy :",acc2) acc_vector = c(acc_vector,acc2)

  #Precision
  pre12<-(cf2[1,1]/sum(cf2[,1]))
  cat("\n    Precision of Class N :",pre12)
  pre22<-(cf2[2,2]/sum(cf2[,2]))
  cat("\n    Precision of Class P :",pre22)
  pre33<-(cf2[3,3]/sum(cf2[,3]))
  cat("\n    Precision of Class Y :",pre33)
  
  prec_n = c(prec_n,pre12)
  prec_p = c(prec_p,pre22)
  prec_y = c(prec_y,pre33)
  
  #Recall
  rec12<-(cf2[1,1]/sum(cf2[1,]))
  cat("\n    Recall of Class N :",rec12)
  rec22<-(cf2[2,2]/sum(cf2[2,]))
  cat("\n    Recall of Class P : ",rec22)
  rec33<-(cf2[3,3]/sum(cf2[3,]))
  cat("\n    Recall of Class Y :",rec33)
  cat("\n")      
  cat("\n")
  cat("\n")
  
  rec_n = c(rec_n,rec12)
  rec_p = c(rec_p,rec22)
  rec_y = c(rec_y,rec33)

Visualization---

  plot(acc_vector,type="o",col="red",xlab="Algorithms",ylab="Accuracy",xaxt='n',lwd=5,cex.lab=1.5)
  axis(1,at=c(1,2,3),labels=c("nn","knn","rf"),cex.axis=2)
  
  plot(prec_n,type="o",col="red",xlab="Algorithms",ylab="Precision",ylim=c(0.7,1.1),xaxt='n',lwd=2,cex.lab=1.5)
  lines(prec_p, type="o",col="blue",lwd=2)
  lines(prec_y, type="o",col="green",lwd=2)
  legend(1,0.92,legend = c("Class N","Class P","Class Y"),fill = c("red","blue","green"),cex=1)
  axis(1,at=c(1,2,3),labels=c("nn","knn","rf"),cex.axis=2)
  
  plot(rec_n,type="o",col="red",xlab="Algorithms",ylab="Recall",ylim=c(0.7,1.1),xaxt='n',lwd=2,cex.lab=1.5)
  lines(rec_p, type="o",col="blue",lwd=2)
  lines(rec_y, type="o",col="green",lwd=3)
  legend(1,0.93,legend = c("Class N","Class P","Class Y"),fill = c("red","blue","green"),cex=1)
  axis(1,at=c(1,2,3),labels=c("nn","knn","rf"),cex.axis=2)
  
  
  
  ###3)With feature selection then Upsampling###
  library(caTools)

f<-read.csv("D:\Excel\Diabetes_1.csv") f<-f[,c(-1,-2)]

#data cleaning---- #remove duplicate values
f<-unique(f)

#Fixing and standardizing typos #1. Gender for(i in 1:length(f$Gender)){ if (f$Gender[i]=='M'){ f$Gender[i]=0; } else{ f$Gender[i]=1; } } #2. CLASS for(i in 1:length(f$CLASS)){ if (f$CLASS[i]=='N'){ f$CLASS[i]=1; } else if(f$CLASS[i]=='P'){ f$CLASS[i]=2; } else{ f$CLASS[i]=3; } } f<-transform(f,Gender=as.integer(Gender),CLASS=as.integer(CLASS));

#feature selection---- library(Boruta) set.seed(1234) imp_features <- Boruta(CLASS~.,data=f,doTrace=2) print(imp_features)

#Removing unimportant features
  #Unimportant features found: Gender, HDL
  f$Gender=NULL
  f$HDL=NULL

#upsampling---- N<-which(f$CLASS==1) P<-which(f$CLASS==2) Y<-which(f$CLASS==3) length(N) length(P) length(Y)

  set.seed(121)
  Up.sampling1<-sample(N,length(Y),replace = TRUE)
  Up.sampling2<-sample(P,length(Y),replace = TRUE)
  f<-f[c(Up.sampling1,Up.sampling2,Y),]

#dividing into training and testing data---- set.seed(123) split<-sample.split(f,SplitRatio = 0.7) tr_data<-subset(f,split==TRUE) ts_data<-subset(f,split==FALSE)

#DECISION TREE---- library(rpart) model2 <- rpart(CLASS ~ . , data=tr_data,method="class")

  #plot(model2, uniform=TRUE, main="Classification Tree for Diabetes")
  #text(model2, use.n=TRUE, all=TRUE, cex=.8)
  
  treePred <- predict(model2, ts_data, type = 'class')
  tb1 = table(treePred, ts_data$CLASS)
  cat("\n")
  cat("\nDecison tree :")

#Accuracy accuracy_Test <- sum(diag(tb1)) / sum(tb1) cat("\n Accuracy :",accuracy_Test) acc_vector = c(accuracy_Test)

#Precision pre91<-(tb1[1,1]/sum(tb1[,1])) cat("\n Precision of Class N :",pre91) pre92<-(tb1[2,2]/sum(tb1[,2])) cat("\n Precision of Class P :",pre92) pre93<-(tb1[3,3]/sum(tb1[,3])) cat("\n Precision of Class Y :",pre93)

prec_n = c(pre91) prec_p = c(pre92) prec_y = c(pre93)

  #Recall
  rec91<-(tb1[1,1]/sum(tb1[1,]))
  cat("\n    Recall of Class N :",rec91)
  rec92<-(tb1[2,2]/sum(tb1[2,]))
  cat("\n    Recall of Class P :",rec92)
  rec93<-(tb1[3,3]/sum(tb1[3,]))
  cat("\n    Recall of Class Y :",rec93,"\n")
  
  rec_n = c(rec91)
  rec_p = c(rec92)
  rec_y = c(rec93)

#KNN----

  tr1_data = tr_data
  ts1_data = ts_data

#Normalize tr1_data[,1:9]<-scale(tr1_data[,1:9]) ts1_data[,1:9]<-scale(ts1_data[,1:9])

#Apply knn library(class) y_pred<-knn(train = tr1_data[-10], test = ts1_data[-10], cl = tr1_data[,10], k = 3)

#Confusion Matrix cf1<-table(actual=ts1_data$CLASS,predicted = y_pred)

  cat("\nKNN :")
  #accuracy
  acc<-(cf1[[1,1]]+cf1[[2,2]]+cf1[[3,3]])/sum(cf1)
  cat("\n    Accuracy :",acc)
  acc_vector = c(acc_vector,acc)
  
  #Precision
  pre1<-(cf1[1,1]/sum(cf1[,1]))
  cat("\n    Precision of Class N :",pre1)
  pre2<-(cf1[2,2]/sum(cf1[,2]))
  cat("\n    Precision of Class P :",pre2)
  pre3<-(cf1[3,3]/sum(cf1[,3]))
  cat("\n    Precision of Class Y :",pre3)
  
  prec_n = c(prec_n,pre1)
  prec_p = c(prec_p,pre2)
  prec_y = c(prec_y,pre3)
  
  #Recall
  rec1<-(cf1[1,1]/sum(cf1[1,]))
  cat("\n    Recall of Class N :",rec1)
  rec2<-(cf1[2,2]/sum(cf1[2,]))
  cat("\n    Recall of Class P :",rec2)
  rec3<-(cf1[3,3]/sum(cf1[3,]))
  cat("\n    Recall of Class Y :",rec3,"\n")
  
  rec_n = c(rec_n,rec1)
  rec_p = c(rec_p,rec2)
  rec_y = c(rec_y,rec3)

#RANDOM fOREST---- library("stats") library("dplyr") library("randomForest")

  ts2_data<-ts_data
  tr2_data<-tr_data
  set.seed(345)

#Applying random forest rfm= randomForest(CLASS~.,data = tr2_data,ntree=10) y_pred1= predict(rfm,ts2_data)

  y_pred1[y_pred1<1.5]=1
  y_pred1[1.5<=y_pred1 & y_pred1<2.5]=2
  y_pred1[2.5<=y_pred1]=3

  ts2_data$y_pred1 = y_pred1

#Confusion Matrix cf2<-table(actual=ts2_data$CLASS,predicted = y_pred1)

  cat("\nRANDOM FOREST:")

#accuracy acc2<-(cf2[[1,1]]+cf2[[2,2]]+cf2[[3,3]])/sum(cf2) cat("\n Accuracy :",acc2) acc_vector = c(acc_vector,acc2)

  #Precision
  pre12<-(cf2[1,1]/sum(cf2[,1]))
  cat("\n    Precision of Class N :",pre12)
  pre22<-(cf2[2,2]/sum(cf2[,2]))
  cat("\n    Precision of Class P :",pre22)
  pre33<-(cf2[3,3]/sum(cf2[,3]))
  cat("\n    Precision of Class Y :",pre33)
  
  prec_n = c(prec_n,pre12)
  prec_p = c(prec_p,pre22)
  prec_y = c(prec_y,pre33)
  
  #Recall
  rec12<-(cf2[1,1]/sum(cf2[1,]))
  cat("\n    Recall of Class N :",rec12)
  rec22<-(cf2[2,2]/sum(cf2[2,]))
  cat("\n    Recall of Class P : ",rec22)
  rec33<-(cf2[3,3]/sum(cf2[3,]))
  cat("\n    Recall of Class Y :",rec33)
  cat("\n")      
  cat("\n")
  cat("\n")
  
  rec_n = c(rec_n,rec12)
  rec_p = c(rec_p,rec22)
  rec_y = c(rec_y,rec33)

Visualization---

  plot(acc_vector,type="o",col="red",xlab="Algorithms",ylab="Accuracy",xaxt='n',lwd=5,cex.lab=1.5)
  axis(1,at=c(1,2,3),labels=c("nn","knn","rf"),cex.axis=2)
  
  plot(prec_n,type="o",col="red",xlab="Algorithms",ylab="Precision",ylim=c(0.7,1.1),xaxt='n',lwd=5,cex.lab=1.5)
  lines(prec_p, type="o",col="blue",lwd=5)
  lines(prec_y, type="o",col="green",lwd=5)
  legend(1,0.92,legend = c("Class N","Class P","Class Y"),fill = c("red","blue","green"),cex=1)
  axis(1,at=c(1,2,3),labels=c("nn","knn","rf"),cex.axis=2)
  
  plot(rec_n,type="o",col="red",xlab="Algorithms",ylab="Recall",ylim=c(0.7,1.1),xaxt='n',lwd=5,cex.lab=1.5)
  lines(rec_p, type="o",col="blue",lwd=5)
  lines(rec_y, type="o",col="green",lwd=5)
  legend(1,0.93,legend = c("Class N","Class P","Class Y"),fill = c("red","blue","green"),cex=1)
  axis(1,at=c(1,2,3),labels=c("nn","knn","rf"),cex.axis=2)

About

Identified the optimal order of preprocessing for Pima Diabetes Dataset, how feature selection and upsampling affects the evaluation metrics

Topics

sampling data-preprocessing data-cleaning feature-selection-methods diabtetes-prediction

Resources

Readme
Activity

Stars

0 stars

Watchers

1 watching

Forks

0 forks
Report repository

Releases

No releases published

Packages

No packages published