knn.Rmd

---
title: "knn"
author: "zilani"
date: "23/07/2019"

output: 
    html_document:
      theme: flatly 
      highlight: tango
      toc: true
      toc_float: true
---

<style>
h3 {
  color: white;
  background-color: #44546a;
  text-indent: auto; 
</style>


```{r setup, include=FALSE}
knitr::opts_chunk$set(echo = TRUE)
```

# Library
```{r}
library(ISLR)
library(MASS)
library(caret)
set.seed(1)
source("functions_w6.R")
library(e1071)
library(mlbench)
library(class)

```

**#########################################################################################**

# knn smarket data 

**### test train split and build the model ----------------------------------------------**


```{r}
# using the smarket data from ISLR LIBRARY for knn classification
# here the target variable is direction to determine whether price is moving up/down
head(Smarket)
```

**we will use the test train split of the data based on the year of the stock price. since here we have 5 years of data we use the data upto 2004 as the training data and the rest mean 2005 data is being the test sample or holdout sample**

```{r}

set.seed(1)
train<-(Smarket$Year<2005)

# use the first two columns to explain the direction of smarket using cbind
trainsmarket  <- cbind(Smarket$Lag1,Smarket$Lag2)[train,]
testsmarket <- cbind(Smarket$Lag1,Smarket$Lag2)[-train,]

# see the dimention of train and test
dim(trainsmarket)
dim(testsmarket)

#making the class based on the train data and also see the length of it which has to be the same lenght of train data

smarket_class_train=Smarket$Direction[train]
length(smarket_class_train)

#  build the knn model
pred=knn(trainsmarket, testsmarket, smarket_class_train,k=3)
```

```{r}

# prediction we made before is been checked with the 
# actual value ofdirection.2005

direction.2005<-Smarket$Direction[-train]
table(pred,direction.2005)
```


```{r}
# checking the accuricy where the accuricy is
# 89.91%

mean(pred==direction.2005)*100
```

**#### plot the optimum k value with the accuricy curv (smarket)**

```{r}
# what is the optimam k value
# using k from 1 to 20 see the accuricy
predicted.purchase = NULL
accuricy.rate = NULL

for(i in 1:20){
    set.seed(1)
    smarketpred = knn(trainsmarket,testsmarket,smarket_class_train,k=i)
    accuricy.rate[i] = mean(smarketpred==direction.2005)
}

# print the rate of accuricy for various k value
#put that in to a data frame
# print the data frame
print(accuricy.rate)
k.values <- 1:20
accuricy.df <- data.frame(accuricy.rate,k.values)

accuricy.df
```

**#### PLOT THE CURB ----------------------**
```{r}
# plot the accuricy rate iris
ggplot(accuricy.df,aes(x=k.values,y=accuricy.rate)) + geom_point()+ geom_line(lty="dotted",color='red')
```



**#########################################################################################**

# knn caravan data 

**#### ISLR data**

**### test train split and build the model ----------------------------------------------**


**this data is in ISLR package where the target variable is Parses?? yes or no mean the customes did bought the policy or not**

**#### data preparation (caravan)--------------------------------**

```{r}
# structure of data file
#str(Caravan)

# summary of target variable
summary(Caravan$Purchase)

# check the null value
any(is.na(Caravan))
```

**Because the KNN classifier predicts the class of a given test observation by identifying the observations that are nearest to it, the scale of the variables matters. Any variables that are on a large scale will have a much larger effect on the distance between the observations, and hence on the KNN classifier, than variables that are on a small scale.**

```{r}
# example there is a variance in two variables not standard
# so need 
var(Caravan[,1])
var(Caravan[,2])
```

```{r}
# save the Purchase column in a separate variable
purchase <- Caravan[,86]

# Standarize the dataset using "scale()" R function
# using all the columns but the last/target columns
standardized.Caravan <- scale(Caravan[,-86])
```

```{r}
# lets check again
var(standardized.Caravan[,1])
var(standardized.Caravan[,2])
```

**### test train with knn (caravan) -------------------------------------------------**

```{r}
# another way of test train split
# First 1000 rows for test set
test.index <- 1:1000
caravantest <- standardized.Caravan[test.index,]
caravan_test_class <- purchase[test.index]

# Rest of data for training
caravantrain <- standardized.Caravan[-test.index,]
caravan_train_class <- purchase[-test.index]

# building knn
set.seed(1)
caravanpred <- knn(caravantrain,caravantest,caravan_train_class,k=5)

# confusion matrix on test data class
table(caravanpred,caravan_test_class)

# checking the accuricy where the accuricy is
# 88.2% good accuricy

mean(caravanpred==caravan_test_class)*100

```

**### plot the optimum k value with the accuricy curv (caravan)-------------------------------**

```{r}
# what is the optimam k value
# using k from 1 to 20 see the accuricy
predicted.purchase = NULL
accuricy.rate = NULL

for(i in 1:20){
    set.seed(1)
    caravanpred = knn(caravantrain,caravantest,caravan_train_class,k=i)
    accuricy.rate[i] = mean(caravanpred==caravan_test_class)
}

# print the rate of accuricy for various k value
#put that in to a data frame
# print the data frame
print(accuricy.rate)
k.values <- 1:20
accuricy.df <- data.frame(accuricy.rate,k.values)

accuricy.df
```

```{r}
# plot the accuricy rate
ggplot(accuricy.df,aes(x=k.values,y=accuricy.rate)) + geom_point()+ geom_line(lty="dotted",color='red')
```


**#########################################################################################**

# knn iris data 

**### test train split and build the model ----------------------------------------------**



```{r}
# spliting the iris data 70/30 split
set.seed(1)

library(caTools)

sample <- sample.split(iris$Species, SplitRatio = .70)
train <- subset(iris, sample == TRUE)
test <- subset(iris, sample == FALSE)

# length and dim of some important things
dim(train)
dim(test)
length(train$Species)

# predection of species in test data
prediris <- knn(train[1:4],test[1:4],train$Species,k=4)

# confusion matrix on test data class
table(prediris,test$Species)

# checking the accuricy where the accuricy is
# 92% good accuricy
accuricy<-mean(prediris==test$Species)*100
print(paste('Accuricy is-----', accuricy, '%'))

```



**### plot the optimum k value with the accuricy curv (iris method-2) ------------**

```{r}
# what is the optimam k value
# using k from 1 to 20 see the accuricy
predicted.purchase = NULL
accuricy.rate = NULL

for(i in 1:20){
    set.seed(1)
    irispred = knn(train[1:4],test[1:4],train$Species,k=i)
    accuricy.rate[i] = mean(irispred==test$Species)
}

# print the rate of accuricy for various k value
#put that in to a data frame
# print the data frame
print(accuricy.rate)
k.values <- 1:20
accuricy.df <- data.frame(accuricy.rate,k.values)

accuricy.df
```

```{r}
# plot the accuricy rate iris
ggplot(accuricy.df,aes(x=k.values,y=accuricy.rate)) + geom_point()+ geom_line(lty="dotted",color='red')
```

**#########################################################################################**

# knn (Sonar) data

**#### using 10 fold cross Validation ----------------------------------------------**


```{r}
data(Sonar)
dim(Sonar)

# create cross validation folds
library(caret)
set.seed(1)
fold <- createFolds(Sonar$Class, k=10)

# apply 10-fold cross-validation
knn.TP <- knn.TN <- knn.FP <- knn.FN <- c()
lda.TP <- lda.TN <- lda.FP <- lda.FN <- c()
glm.TP <- glm.TN <- glm.FP <- glm.FN <- c()
svm.line.TP <- svm.line.TN <- svm.line.FP <- svm.line.FN <- c()

for(i in 1:length(fold)){
    # true label for fold i
    truth <- Sonar$Class[fold[[i]]]

    # apply knn for classification
    preds <- knn(Sonar[-fold[[i]],-61], Sonar[fold[[i]],-61], Sonar$Class[-fold[[i]]], k=5)
    knn.TP <- c(knn.TP, sum((truth == preds)[truth == "M"]))
    knn.TN <- c(knn.TN, sum((truth == preds)[truth == "R"]))
    knn.FP <- c(knn.FP, sum((truth != preds)[truth == "R"]))
    knn.FN <- c(knn.FN, sum((truth != preds)[truth == "M"]))
    
    
}
print("knn evaluation")
evaluate(knn.TN, knn.FP, knn.TP, knn.FN)
```



**#########################################################################################**

# knn (Ionosphere) data

**#### using 10 fold cross Validation ----------------------------------------------**


```{r}
data(Ionosphere)
df <- Ionosphere
df <- df[,-2]
df <- df[,-1]


# apply 10-fold cross-validation
fold <- createFolds(df$Class, k=10)
knn.TP <- knn.TN <- knn.FP <- knn.FN <- c()


for(i in 1:length(fold)){
    # true label for fold i
    truth <- df$Class[fold[[i]]]

    # apply knn for classification
    preds <- knn(df[-fold[[i]],-33], df[fold[[i]],-33], df$Class[-fold[[i]]], k=5)
    knn.TP <- c(knn.TP, sum((truth == preds)[truth == "good"]))
    knn.TN <- c(knn.TN, sum((truth == preds)[truth == "bad"]))
    knn.FP <- c(knn.FP, sum((truth != preds)[truth == "bad"]))
    knn.FN <- c(knn.FN, sum((truth != preds)[truth == "good"]))
    
    
}
print("knn evaluation")
evaluate(knn.TN, knn.FP, knn.TP, knn.FN)
```