Analyzing K-Nearest Neighbors Model Accuracy with

School: Georgia Institute Of Technology - Course: ISYE 6501 - Subject: Accounting

HOMEWORK WEEK 2 2023-01-23 Question 3.1 #options("scipen"=100, "digits"=10) #install.packages("rmarkdown") #Get datatset to a dataframe df path2<-file.path("/Users/~/Documents/ISYE 6501/Homework/Week 1/data 2.2/credit_card_data.txt") data2<-read.table(path2,header =FALSE) print(dim(data2))#check size of dataset of credit_card_data.txt ## [1] 65411 head(data2) ##V1V2V3V4 V5 V6 V7 V8V9 V10 V11 ## 11 30.83 0.000 1.251011 20201 ## 20 58.67 4.460 3.04106143 5601 ## 30 24.50 0.500 1.501101 280 8241 ## 41 27.83 1.540 3.751050 10031 ## 51 20.17 5.625 1.711101 12001 ## 61 32.08 4.000 2.501100 36001 Question 3.1.a Train Knn with k-fold (cv.kknn) cross-validation #Use credit_card_data.txt #install.packages("kknn") #Train Knn with k-fold (cv.kknn) cross-validation #Ref. document: https://www.rdocumentation.org/packages/kknn/versions/1.3.1/topics/tra in.kknn data=data.frame(data2) library(kknn) set.seed(123) knn_model_cv_acc_vector=c() for(Kin1:50){#loop throuh 50 values of K to check the accuracy of knn model trained with 10-fold cross validation knn_model_cv=cv.kknn(V11~., data,kcv =10,k =K,kernel = "optimal",scale =T) knn_model_cv_result=data.frame(knn_model_cv[1]) knn_pred=round(knn_model_cv_result[,2])

knn_model_cv_acc=sum(knn_pred==data[,11])/nrow(data) knn_model_cv_acc_vector<-c(knn_model_cv_acc_vector, knn_model_cv_acc) } knn_model_cv_acc_vector ##[1] 0.8119266 0.8073394 0.8134557 0.8103976 0.8547401 0.8455657 0.8455657 ##[8] 0.8425076 0.8425076 0.8409786 0.8455657 0.8470948 0.8440367 0.8547401 ## [15] 0.8516820 0.8470948 0.8440367 0.8425076 0.8455657 0.8516820 0.8425076 ## [22] 0.8425076 0.8425076 0.8394495 0.8440367 0.8425076 0.8425076 0.8425076 ## [29] 0.8363914 0.8394495 0.8440367 0.8425076 0.8455657 0.8363914 0.8348624 ## [36] 0.8440367 0.8302752 0.8348624 0.8302752 0.8287462 0.8409786 0.8333333 ## [43] 0.8440367 0.8501529 0.8409786 0.8363914 0.8318043 0.8455657 0.8379205 ## [50] 0.8409786 #Look for the best accuracy and respective K values max(knn_model_cv_acc_vector) ## [1] 0.8547401 which(knn_model_cv_acc_vector==max(knn_model_cv_acc_vector)) ## [1]5 14 #Look for the worst accuracy and respective K values min(knn_model_cv_acc_vector) ## [1] 0.8073394 which(knn_model_cv_acc_vector==min(knn_model_cv_acc_vector)) ## [1] 2 As shown above, with dataset of credit_card_data.txt and 10-fold cross validation, the proposed KNN model works well with K of 5 or 14 with the accuracy of about 85.47%, while the small K value of 2 maintains the model's acuracy at a minimum rate of about 80%.