Data Pre-Processing

# Import iris dataset
url <- "http://archive.ics.uci.edu/ml/machine-learning-databases/iris/iris.data"
iris <- read.csv(url)

#see dimension
dim(iris)

# see row number
nrow(iris)

# see column number
ncol(iris)

#see row names
names(iris)

#see column names
colnames(iris)

#see the first few data
head(iris)
head(iris, 10)

#see structure
str(iris)

# see summary
summary(iris)

#change the column name
colnames(iris) <- c("Sepal.Length","Sepal.Width","Petal.Length", "Petal.Width","Species")
colnames(iris)

#get percentage of missing value of the attributes - Approach 2 (Function)
sapply(iris, function(df)
{
  sum(is.na(df)==T)/length(df)
})


##################################################################################################################
#Titanic Dataset
# Import the training set: train
?read.csv
train <- read.csv (file.choose(), stringsAsFactors = F,na.strings=c("","NA"," "))
test <- read.csv (file.choose(), stringsAsFactors = F,na.strings=c("","NA"," "))

str(train)
str(test)
summary(train)
summary(test)

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

#Titanic Dataset
# Import the training set: train
?read.csv
train <- read.csv (file.choose(), stringsAsFactors = F,na.strings=c("","NA"," "))
test <- read.csv (file.choose(), stringsAsFactors = F,na.strings=c("","NA"," "))

str(train)
str(test)
summary(train)
summary(test)

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

#Data Type conversion
train$Survived = as.factor(train$Survived)
train$Pclass = as.factor(train$Pclass)

str(train)
summary(train)

#################################################################################
#detect missing values - Approach 1 (lengthy)
is.na(train$Age)
sum(is.na(train$Age)==TRUE)
sum(is.na(train$Age)==TRUE)/length(train$Age)
sum(is.na(train$Fare)==TRUE)/length(train$Fare)

#get percentage of missing value of the attributes - Approach 2 (Function)
sapply(train, function(df)
{
  sum(is.na(df)==T)/length(df)
})

#Approach - Amelia Package
#install.packages("Amelia")
library("Amelia")

missmap(train, main = "Missing Map")
AmeliaView()

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


#get percentage of missing value of the attributes - Approach 2 (Function)
sapply(train, function(df)
{
  sum(is.na(df)==T)/length(df)
})

#Approach - Amelia Package
#install.packages("Amelia")
library("Amelia")

missmap(train, main = "Missing Map")
AmeliaView()

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

# Imputing Missing Value

# Substitute the missing values with the average value
sum(is.na(train$Age)==TRUE)/length(train$Age)
train$Age[is.na(train$Age)] <- mean(train$Age,na.rm=T)
sum(is.na(train$Age))

# Missing Value Imputation - Embarked
table(train$Embarked, useNA = "always")
#Mode = S

# Substitute the missing values with the mode value
train$Embarked[is.na(train$Embarked)] <- 'S'
sum(is.na(train$Embarked))
table(train$Embarked, useNA = "always")


# Missing Value Imputation - Fare
# Substitute the missing values with the average value
train$Fare[is.na(train$Fare)] <- mean(train$Fare,na.rm=T)
sum(is.na(train$Fare))

#Check again for NA
sapply(train, function(df)
{
  sum(is.na(df)==T)/length(df)
})
##################################################################################################################


##univariate EDA

#categorical variables
titanic_train <- train
xtabs(~Survived,titanic_train)
summary(titanic_train$Survived)
ggplot(titanic_train) + geom_bar(aes(x=Survived))

summary(titanic_train$Sex)
ggplot(titanic_train) + geom_bar(aes(x=Sex))

summary(titanic_train$Pclass)
ggplot(titanic_train) + geom_bar(aes(x=Pclass))

#numerical variables
summary(titanic_train$Fare)
ggplot(titanic_train) + geom_histogram(aes(x=Fare),fill = "white", colour = "black")
ggplot(titanic_train) + geom_boxplot(aes(x=factor(0),y=Fare)) + coord_flip()
ggplot(titanic_train) + geom_density(aes(x=Fare))

summary(titanic_train$Age)
ggplot(titanic_train) + geom_histogram(aes(x=Age),fill = "white", colour = "black")
ggplot(titanic_train) + geom_boxplot(aes(x=factor(0),y=Age)) + coord_flip()

#####################################################################################
##bivariate EDA
#C-C relationships
xtabs(~Survived+Sex,titanic_train)
ggplot(titanic_train) + geom_bar(aes(x=Sex, fill=factor(Survived)))

xtabs(~Survived+Pclass,titanic_train)
ggplot(titanic_train) + geom_bar(aes(x=Pclass, fill=factor(Survived)) )

xtabs(~Survived+Embarked,titanic_train)
ggplot(titanic_train) + geom_bar(aes(x=Embarked, fill=factor(Survived)) )

#N-C relationships
ggplot(titanic_train) + geom_boxplot(aes(x = factor(Survived), y = Age))
ggplot(titanic_train) + geom_histogram(aes(x = Age),fill = "white", colour = "black") + facet_grid(factor(Survived) ~ .)

ggplot(titanic_train) + geom_boxplot(aes(x = factor(Survived), y = Fare))
ggplot(titanic_train) + geom_histogram(aes(x = Fare),fill = "white", colour = "black") + facet_grid(factor(Survived) ~ .)

#####################################################################################
##multivariate EDA
xtabs(~factor(Survived)+Pclass+Sex,titanic_train)
ggplot(titanic_train) + geom_bar(aes(x=Sex, fill=factor(Survived))) + facet_grid(Pclass ~ .)


xtabs(~Survived+Embarked+Sex,titanic_train)
ggplot(titanic_train) + geom_bar(aes(x=Sex, fill=factor(Survived))) + facet_grid(Embarked ~ .)
#####################################################################################
#EDA - Exploratory Data Analysis
#One by One variable approach
library(plyr)
library(rpart)
library(caret)
library(caTools)
library(mice)
library(stringr)
library(Hmisc)
library(ggplot2)
library(vcd)
library(ROCR)
library(pROC)
library(VIM)
library(glmnet)
library(ggmosaic)

#########################################################################
# Feature Engineering
#########################################################################
#Combine all the data. Before combining we need to add Survived column in test dataset
test$Survived <- NA
Full <- rbind(train,test)
str(Full)

# Engineered variable 1: Title
# Create the column child, and indicate whether child or no child
Full$Child <- NA
Full$Child[Full$Age < 18] <- 1
Full$Child[Full$Age >= 18] <- 0
str(Full$Child)


# Engineered variable 2: Title
# Extract the title - Mr, Mrs, Miss
Full$Title <- sapply(Full$Name, FUN=function(x) {strsplit(x, split='[,.]')[[1]][2]})
Full$Title <- sub(' ', '', Full$Title)  # Remove the white space or blank
table(Full$Title)
barplot(table(Full$Title))
?barplot

# Combine small title groups
Full$Title[Full$Title %in% c('Mme', 'Mlle')] <- 'Mlle'
Full$Title[Full$Title %in% c('Capt', 'Don', 'Major', 'Sir')] <- 'Sir'
Full$Title[Full$Title %in% c('Dona', 'Lady', 'the Countess', 'Jonkheer')] <- 'Lady'
# Convert to a factor
Full$Title <- factor(Full$Title)
table(Full$Title)
barplot(table(Full$Title))


# Engineered variable 3: Family size
Full$FamilySize <- Full$SibSp + Full$Parch + 1
table(Full$FamilySize)

# Split back into test and train sets
train_Featured <- Full[1:891,]
test_Featured <- Full[892:1309,]

train_Featured$Survived <- as.factor(train_Featured$Survived)
train_Featured$Sex <- as.factor(train_Featured$Sex)
train_Featured$Embarked <- as.factor(train_Featured$Embarked)

test_Featured$Sex <- as.factor(test_Featured$Sex)
test_Featured$Embarked <- as.factor(test_Featured$Embarked)

# Build Random Forest Ensemble
set.seed(415)
library("randomForest")
fit <- randomForest(as.factor(Survived) ~ Pclass + Sex + Age + SibSp + Parch + Fare + Embarked + Child + Title + FamilySize + FamilyID2,
                    data=train_Featured, importance=TRUE, ntree=2000)
# Look at variable importance
varImpPlot(fit)

Machine learning Data modelling

Feature Engineering:

1.     Adding or dropping feature

a.     Choose the feature that has most signals

2.     Combine multiple feature into one feature

a.     Represent the data in the simplest way possible(like all measurements in feet rather than some in inches

3.     Binning

a.     Replace an exact numerical measurement with a broader category. Like replacing unwanted extra measurement into True or False (like size of pool which is not that important rather having or not having pool is important)

4.     One-hot encoding

a.     A way to represent categorical data as number without creating

Also it’s advisable to have atleast 10x rows for X number of features