OnlineNewsPopularity.R

##########
#Load data
setwd("~/Dropbox/DePaul/OnlineNewsPopularity")
OnlineNewsPopularity = read.csv("OnlineNewsPopularity.csv", header = TRUE)

###########Data Preprpcessing step##########

#Duplicate dataset
df.onNews <- OnlineNewsPopularity

#Remove ID
df.onNews$url  = NULL
df.onNews$timedelta = NULL

#Change Dummy variables to factors
df.onNews$pubDay <- ifelse(OnlineNewsPopularity$weekday_is_monday == 1, "monday", 
                           ifelse((OnlineNewsPopularity$weekday_is_tuesday == 1), "tuesday", 
                                  ifelse((OnlineNewsPopularity$weekday_is_wednesday == 1), "wednesday", 
                                         ifelse((OnlineNewsPopularity$weekday_is_thursday == 1), "thursday", 
                                                ifelse((OnlineNewsPopularity$weekday_is_friday == 1), "friday", 
                                                       ifelse((OnlineNewsPopularity$weekday_is_saturday == 1), "saturday", "sunday"))))))
df.onNews$pubDay <- as.factor(df.onNews$pubDay)


df.onNews$weekday_is_monday    = NULL
df.onNews$weekday_is_tuesday   = NULL
df.onNews$weekday_is_wednesday = NULL
df.onNews$weekday_is_thursday  = NULL
df.onNews$weekday_is_friday    = NULL
df.onNews$weekday_is_saturday  = NULL
df.onNews$weekday_is_sunday    = NULL


df.onNews$dataChannel <- ifelse(OnlineNewsPopularity$data_channel_is_lifestyle == 1,"lifestyle",
                                ifelse(OnlineNewsPopularity$data_channel_is_entertainment == 1,"entertainment",
                                       ifelse(OnlineNewsPopularity$data_channel_is_bus == 1,"business",
                                              ifelse(OnlineNewsPopularity$data_channel_is_socmed == 1,"Social",
                                                     ifelse(OnlineNewsPopularity$data_channel_is_tech == 1,"tech", "world")))))

df.onNews$data_channel_is_lifestyle = NULL
df.onNews$data_channel_is_entertainment = NULL
df.onNews$data_channel_is_bus = NULL
df.onNews$data_channel_is_socmed = NULL
df.onNews$data_channel_is_tech = NULL
df.onNews$data_channel_is_world = NULL

df.onNews$dataChannel <- as.factor(df.onNews$dataChannel)


#Class distribution Bin data
df.onNews$share2bins <- ifelse(OnlineNewsPopularity$shares > 1300, "high", "low")
df.onNews$share2bins <- as.factor(df.onNews$share2bins)


######
library(ggplot2)
theme_set(theme_bw(base_size = 28))

qplot(df.onNews$dataChannel, geom="histogram",ylab = "count",binwidth = 30,
      main = "Histogram for data channel", xlab = "Channel", fill=df.onNews$share2bins ) 

table(df.onNews$dataChannel,df.onNews$share2bins)

qplot(df.onNews$pubDay, geom="histogram",ylab = "count",binwidth = 30,
      main = "Histogram for publication day", xlab = "Day", fill=df.onNews$share2bins ) 

table(df.onNews$pubDay,df.onNews$share2bins)

qplot(df.onNews$self_reference_min_shares, geom="histogram", ylab = "count",
      main = "Histogram for self_reference_min_shares", xlab = "Min. shares of referenced articles in Mashable", fill=df.onNews$share2bins, xlim=c(3.8,8000) ) 

table(df.onNews$self_reference_min_shares,df.onNews$share2bins0)

qplot(df.onNews$kw_avg_avg, geom="histogram",ylab = "count",
      main = "Histogram for kw_avg_avg", xlab = "Avg. keyword", fill=df.onNews$share2bins, xlim=c(3.8,8000) ) 

hist(df.onNews$shares, xlim=c(3.8,500000) )


#######
df.onNews$shares = NULL

###########Required Library##########
library(rpart.plot)
library(caret)
library(e1071)
library(klaR)
library(rattle)
library(rpart.plot)
library(RColorBrewer)

###########Clasification modelling step##########
set.seed(415)
# define an 66%/34% train/test split of the dataset
trainIndex <- createDataPartition(df.onNews$share2bins, p=0.66, list=FALSE)
data_train <- df.onNews[ trainIndex,]
data_test <- df.onNews[-trainIndex,]



############# Decision Tree #####################
Sys.time()

fit.rpart5= train(share2bins ~ . ,data=data_train , method= "rpart", tuneLength = 5, trControl=trainControl(method="cv",number=10))
fit.rpart8= train(share2bins ~ . ,data=data_train , method= "rpart", tuneLength = 8, trControl=trainControl(method="cv",number=10))
fit.rpart13= train(share2bins ~ . ,data=data_train , method= "rpart", tuneLength = 13, trControl=trainControl(method="cv",number=10))
fit.rpart17= train(share2bins ~ . ,data=data_train , method= "rpart", tuneLength = 17, trControl=trainControl(method="cv",number=10))
fit.rpart20= train(share2bins ~ . ,data=data_train , method= "rpart", tuneLength = 20, trControl=trainControl(method="cv",number=10))

summary(fit.rpart5$finalModel)

predictions <- predict(fit.rpart8, data_test[,1:48])
confusionMatrix(predictions, data_test$share2bins)

fancyRpartPlot(fit.rpart8$finalModel)
prp(fit.rpart5$finalModel)
print(fit.rpart8)
print(fit.rpart8$finalModel)
Sys.time()
##################################################
############# Naive Bayes #####################
Sys.time()
fit.nb= train(share2bins ~ . ,data=data_train , method= "nb", trControl=trainControl(method="cv",number=10))

predictions <- predict(fit.nb, data_test[,1:58])

confusionMatrix(predictions, data_test$share2bins)

print(fit.nb$finalModel)
print(fit.nb)
Sys.time()
##################################################
############# Random Forest #####################

fit.rf = train(share2bins ~ . ,data_train , method= "rf", ntree=501 , tuneGrid = data.frame(mtry = 4), 
               allowParallel=TRUE,trControl=trainControl(method="cv",number=10) )


predictions <- predict(fit.rf2, data_test[,1:48])
confusionMatrix(predictions, data_test$share2bins)

print(fit.rf)
print(fit.rf$finalModel)
importance(fit.rf$finalModel)
varImpPlot(fit.rf2$finalModel)
importance(fit.rf$finalModel)

##################################################