siner

Artifical Intelligence Model using Machine learning Algorithm to predict the ICO ( Cryptocurrencies ) offerings for Fundraising teams and Startups

rm(list=ls()) icobench<- read.csv("Downloads/Machinelearningcourseworkright.csv") View(icobench) str(icobench) icobench$success<-factor(icobench$success) str(icobench$success) sum(complete.cases(icobench)) sum(!complete.cases(icobench)) summary(icobench) View(icobench)

install.packages("VIM") library("VIM") aggr(icobench,numbers=TRUE,prop=FALSE) qicobench<-icobench[which(icobench$hasVideo==1&icobench$rating>=4.0),] hist(icobench$priceUSD) si_icobench<-icobench si_icobench$priceUSD[is.na(si_icobench$priceUSD)]<-mean(si_icobench$priceUSD,na.rm = TRUE) si_icobench hist(icobench$teamSize) si_icobench$teamSize[is.na(si_icobench$teamSize)]<-mean(si_icobench$teamSize,na.rm = TRUE)

si_icobench

str(icobench$success) si_icobench$success[is.na(si_icobench$success)]<-mean(si_icobench$success,na.rm=TRUE) View(si_icobench)

crowdfunding<-si_icobench View(crowdfunding) sum(complete.cases(si_icobench)) sum(!complete.cases(crowdfunding)) dup_crowfunding<-duplicated(crowdfunding$hasVideo) x<-si_icobench x<-x[-which(x$success%in%outliers),] table(x$success,x$minInvestment) table(data$success,data$hasReddit) boxplot(x$priceUSD,plot = FALSE)$out # priceUSD,teamsize,coinnum,dirtubutedpercentage are the outliers outlier<-boxplot(x$priceUSD,plot = FALSE)$out length(x$priceUSD) length(x$teamSize) boxplot(x$priceUSD) sum(complete.cases(x)) sum(!complete.cases(x)) outliers<-boxplot(x$teamSize,plot = FALSE)$out y<-x y<-y[-which(y$teamSize %in% outliers),] length(y) length(y$priceUSD) out<-boxplot(y$coinNum,plot = FALSE)$out z<-y z<-z[-which(z$coinNum %in% out),] length(z$teamSize) sum(!complete.cases(z)) outl<-boxplot(z$distributedPercentage,plot = FALSE)$out w<-z w<-w[-which(w$distributedPercentage %in% outl),] sum(complete.cases(w))

KNN CLASSIFICATION TECHNIQUE

crowdfunding<-crowdfunding[-1]#Removing the id factor table(crowdfunding$success) crowdfunding$success<-as.character(crowdfunding$) View(crowdfunding) round(prop.table(table(crowdfunding$success))*100 , digits = 1) summary(crowdfunding[c("rating","priceUSD","teamSize")]) normalize <- function(x) { if(is.numeric(x)){ x= ((x - min(x)) / (max(x) - min(x))) } return(x) } crowdfunding_n<-as.data.frame(lapply(crowdfunding[2:10],normalize)) summary(crowdfunding_n$priceUSD) crowdfunding_train<-crowdfunding_n[1:2075,] crowdfunding_test<-crowdfunding_n[2076:2767,] crowdfunding_train_labels<-crowdfunding[1:2075,1] crowdfunding_test_labels<-crowdfunding[2076:2767,1] crowdfunding_train_labels install.packages("class") library(class)

We use knn() function to perform classification

We split our data into training and test datasets,

each with exactly the same numeric features.

The labels for the training data are stored

in a separate factor vector.

The only remaining parameter is k,

which specifies the number of neighbors to include in the vote.

K=45

training size =2666 so we try its square root 51 as the value of k first

#Using an odd number of K will reduce the chance of ending with a tie vote. crowdfunding_test_pred<-knn(train =crowdfunding_train,test = crowdfunding_test, cl=crowdfunding_train_labels, k=K ) #error while specifying the crowdfunding_train_labels install.packages("gmodels") library(gmodels) CrossTable(crowdfunding_test_labels,crowdfunding_test_pred, prop.chisq = FALSE, prop.t = FALSE, prop.r = FALSE, dnn = c('predicted', 'actua l')) (542+3)/692

install.packages("Metrics") library(Metrics) mae(crowdfunding_test_pred,crowdfunding_test) library(ROCR) icobench_test_prob<-predict(crowdfunding_test_pred,crowdfunding_test,type = "raw")

options(scipen=999) head(icobench_test_prob) icobench_results<-data.frame(actual_type=crowdfunding_test_labels, predict_type=crowdfunding_test_pred, prob_yes=round(crowdfunding_test_prob[,1],5), prob_no=round(crowdfunding_test_prob[,2],5))

head(icobench_results) head(subset(icobench_results,actual_type!=predict_type)) CrossTable(icobench_results$actual_type,icobench_results$predict_type, dnn=c('actual', 'predict'), prop.chisq = FALSE, prop.t = FALSE, prop.r=FALSE)

install.packages("caret", dependencies = TRUE) library(caret) confusionMatrix(icobench_results$predict_type, icobench_results$actual_type, positive = "N") library(ROCR) ROC<-roc(crowdfunding_test_labels,as.numeric(crowdfunding_test_pred),measure = "tpr",x.measure = "fp") plot(ROC) auc(ROC) pred_object<-prediction(icobench_results$prob_yes,icobench_results$actual_type) roc_NB<-performance(pred_object,measure = "tpr",x.measure = "fpr") plot(roc_NB, main="ROC of success of companies for crowdfunding ",col="blue",lwd=2) abline(a = 0, b = 1, lwd = 2, lty = 2) auc_object_NB <- performance(pred_object, measure = "auc") auc_NB <- [email protected][[1]]

Naive Bayes Classifers

rm(list=ls()) icobench<-read.csv("Downloads/MLNB.csv") View(icobench) icobench<-icobench[-1] View(icobench ) str(icobench) icobench$success<-factor(icobench$success) str(icobench$success) table(icobench$success) install.packages("tm") library(tm) icobench_corpus<-VCorpus(VectorSource(icobench$brandSlogan)) print(icobench_corpus) inspect(icobench_corpus[1:2]) as.character(icobench_corpus[1]) lapply(icobench_corpus[1:5], as.character)

clean up the corpus using tm_map()

icobench_corpus_clean<-tm_map(icobench_corpus,content_transformer(tolower))

show the difference between sms_corpus and corpus_clean

as.character(icobench_corpus[[1]]) as.character(icobench_corpus_clean[[1]])

icobench_corpus_clean<-tm_map(icobench_corpus_clean,removeNumbers) icobench_corpus_clean <- tm_map(icobench_corpus_clean, removeWords, stopwords()) # remove stop words icobench_corpus_clean <- tm_map(icobench_corpus_clean, removePunctuation) # remove punctuat install.packages("SnowballC") library(SnowballC) icobench_corpus_clean<-tm_map(icobench_corpus_clean,stemDocument) icobench_corpus_clean<-tm_map(icobench_corpus_clean,stripWhitespace) lapply(icobench_corpus[1:3],as.character) lapply(icobench_corpus_clean[1:3],as.character) icobench_dtm <- DocumentTermMatrix(icobench_corpus_clean) dim(icobench_dtm) inspect(icobench_dtm[1:5, 1:10]) set.seed(123)

ratio=0.85 p_index=round(nrow(icobench_dtm)*ratio) icobench_dtm_train<-icobench_dtm[1:p_index,] icobench_dtm_test<-icobench_dtm[(p_index +1):nrow(icobench_dtm),] icobench_train_labels<-icobench[1:p_index,]$success icobench_test_labels<-icobench[(p_index +1):nrow(icobench_dtm),]$success prop.table(table(icobench_train_labels)) prop.table(table(icobench_test_labels)) findFreqTerms(icobench_dtm,10) icobench_Freq_words<-findFreqTerms(icobench_dtm,10) str(icobench_Freq_words) icobench_dtm_Freq_train<-icobench_dtm_train[,icobench_Freq_words] icobench_dtm_Freq_test<-icobench_dtm_test[,icobench_Freq_words]

naive bayes clssifier is trained on categorical data

convert counts to categorical variable

convert_counts <- function(x) { x <- ifelse(x > 0, "Yes", "No") }

apply() convert_counts() to columns of train/test data

MARGIN=2 means apply function on column; MARGIN=1 means row

search lapply() we used in last session and compare them

icobench_train<-apply(icobench_dtm_Freq_train, MARGIN = 2,convert_counts) icobench_test<-apply(icobench_dtm_Freq_test, MARGIN = 2,convert_counts) install.packages("e1071") library(e1071) icobench_classifier<-naiveBayes(icobench_train,icobench_train_labels) icobench_test_pred<-predict(icobench_classifier,icobench_test) library(gmodels) CrossTable(icobench_test_pred, icobench_test_labels, prop.chisq = FALSE, prop.t = FALSE, prop.r = FALSE, dnn = c('predicted', 'actual')) confusionMatrix(icobench_test_pred,icobench_test_labels,positive = TRUE,dnn = c("predicted","actual")) (278+18)/415

Step 5: Improving model performance ---

icobench_classifier_laplace<-naiveBayes(icobench_train,icobench_train_labels,laplace =1) icobench_test_pred_laplace<-predict(icobench_classifier_laplace,icobench_test) CrossTable(icobench_test_pred_laplace, icobench_test_labels, prop.chisq = FALSE, prop.t = FALSE, prop.r = FALSE, dnn = c('predicted', 'actual')) #EVALUATION icobench_test_prob<-predict(icobench_classifier,icobench_test,type = "raw")

options(scipen=999) head(icobench_test_prob) icobench_results<-data.frame(actual_type=icobench_test_labels, predict_type=as.numeric(icobench_test_pred), prob_yes=round(icobench_test_prob[,1],10), prob_no=round(icobench_test_prob[,2],10))

head(icobench_results) head(subset(icobench_results,actual_type!=predict_type)) CrossTable(icobench_results$actual_type,icobench_results$predict_type, dnn=c('actual', 'predict'), prop.chisq = FALSE, prop.t = FALSE, prop.r=FALSE)

(278+18)/415 install.packages("caret", dependencies = TRUE)

library(caret) confusionMatrix(icobench_results$predict_type, icobench_results$actual_type, positive = "N")

library(pROC) plot(roc((as.numeric(icobench_test_pred),icobench_test_label,direction="<",col="yellow",lwd = 3 ,main="ROC cCurve"))) ROCurve<-roc(icobench_test_labels , as.numeric(icobench_test_pred,measure = "tpr",x.measure = "fp")) plot(ROCurve) auc(ROCurve) pred_object<-prediction(icobench_results$predict_type,icobench_results$actual_type) roc_NB<-performance(pred_object,measure = "tpr",x.measure = "fpr") plot(roc_NB, main="ROC of success of companies for crowdfunding ",col="blue",lwd=2)

abline(a = 0, b = 1, lwd = 2, lty = 2) auc_object_NB <- performance(pred_object, measure = "auc") auc_NB <- [email protected][[1]] auc_NB

#Decision Tree Analysis

rm(list=ls()) icobench<- read.csv("Downloads/MLDT.csv") View(icobench) str(icobench) icobench$success<-factor(icobench$success)

str(icobench$success) sum(complete.cases(icobench)) sum(!complete.cases(icobench)) summary(icobench) install.packages("VIM") library("VIM") aggr(icobench,numbers=TRUE,prop=FALSE) qicobench<-icobench[which(icobench$hasVideo==1&icobench$rating>=4.0),] hist(icobench$priceUSD) si_icobench<-icobench si_icobench$priceUSD[is.na(si_icobench$priceUSD)]<-mean(si_icobench$priceUSD,na.rm = TRUE) si_icobench hist(icobench$teamSize) si_icobench$teamSize[is.na(si_icobench$teamSize)]<-mean(si_icobench$teamSize,na.rm = TRUE) sum(complete.cases(si_icobench)) ICO<-si_icobench view(ICO)

table(ICO$startDate) table(ICO$endDate) summary(ICO$platform) ICO$hasGithub<-factor(ICO$hasGithub)

smp_size<-floor(0.75*nrow(ICO)) sample(11,6) sample(19,5) set.seed(12345) sample(10,5) train_ico<-sample(nrow(ICO),smp_size) crowdfunding_train<-ICO[train_ico,] crowdfunding_test<-ICO[-train_ico,] #-------------------------------------------------- #2. Training a model on the data: basic #-------------------------------------------------- #We will use the C5.0 algorithm in the #C50 package for training our decision tree model.ICO ICO$hasGithub<-factor(ICO$hasGithub) install.packages("C50") library(C50) library(tidyverse) icobench_model<-C5.0(crowdfunding_train[6] ,crowdfunding_train$success) icobench_model

summary(icobench_model) crowdfunding_pred<-predict(icobench_model,crowdfunding_test) install.packages("Metrics") library('Metrics') mae(crowdfunding_pred,crowdfunding_test$success)

#This creates a vector of predicted class values, #which we can compare to the actual class values #using the CrossTable() function in the gmodels package. #Setting the prop.c and prop.r parameters to FALSE #removes the column and row percentages from the table. #The remaining percentage (prop.t) #indicates the proportion of records in the cell out of the total number of records. library(gmodels) CrossTable(crowdfunding_pred,crowdfunding_test$success, prop.chisq = FALSE, prop.c = FALSE, prop.r = FALSE, dnn = c('predicted default', 'actual')) confusionMatrix(crowdfunding_pred,crowdfunding_test$success,positive="Y")

crowdfunding_boost<-C5.0(select(crowdfunding_train[1]) ,crowdfunding_train$success,trials = 10) crowdfunding_boost summary(crowdfunding_boost)

crowdfunding_boost_pred<-predict(crowdfunding_boost,crowdfunding_test) CrossTable(crowdfunding_boost_pred,crowdfunding_test$success, prop.chisq = FALSE, prop.c = FALSE, prop.r = FALSE, dnn = c('predicted default', 'actual'))

Regression Anlysis

rm(list=ls()) icobench<-read.csv("Downloads/Machinelearningcourseworkright.csv") str(icobench) icobench<-icobench[-1] view(icobench) icobench$success<-factor(icobench$success)

str(icobench$success) smp_size<-floor(0.75*nrow(icobench)) set.seed(987) train_ind<-sample(nrow(icobench),smp_size)

icobench_train<-icobench[train_ind,]

icobench_test<-icobench[-train_ind,] library(rpart) m.rpart<-rpart(success~.,data = icobench) summary(m.rpart) install.packages('rpart.plot') library(rpart.plot) rpart.plot(m.rpart, digits =2) rpart.plot(m.rpart, digits = 4, fallen.leaves = TRUE, type = 3, extra = 101) p.rpart<-predict(m.rpart,icobench_test) library(gmodels) CrossTable(p.rpart,icobench_test$success,prop.chisq = FALSE, prop.c = FALSE, prop.r=FALSE ,dnn=c("predicted","actual"))

install.packages("Metrics") library(Metrics) mae(p.rpart,icobench_test$success) rmse(p.rpart,icobench_test$success) baseline<-mean(icobench_train$success) mae(baseline,icobench_test$success) rmse(baseline,icobench_test$success) library(Cubist) library(tidyverse) m.cubist<-cubist(x=select(icobench_train,-success),y=icobench_train$success) summary(m.cubist) p.cubist<-predict(m.cubist,select(icobench_test,-success)) mae(icobench_test$success,p.cubist) rmse(icobench_test$success,p.cubist)

#random Forest Clasifiers install.packages("randomForest") library(randomForest) m.rf<-randomForest(icobench_test,y=icobench_train[1], ntree = 20) p.rf<-predict(m.rf,icobench[train_ind==2,])