Knn.Rmd

---
title: "Knn"
author: "Santosh Shyamala Ramasubramanian"
date: "April 28, 2018"
output: word_document
---

```{r}
library(ggplot2)
library(dendextend)
library(plyr)
library(reshape2)
library(lattice)
library(caret)
library(class)
library(MASS)
```


```{r}
train <- read.csv("D:/Santosh/Tamu Notes/ISEN 613/Project/HAR Dataset working directory/hartrain.csv")
test <-read.csv("D:/Santosh/Tamu Notes/ISEN 613/Project/HAR Dataset working directory/hartest.csv")
```


```{r}
data= rbind(train,test)
```


```{r}
compute.misclass<-function(conf){
    x= ncol(conf)
    y= nrow(conf)
    mis= 0
    clas= 0
    for (i in (1:x)){
        for (j in (1:y)){
            if(i==j){
                clas= clas+conf[i,j]
            }
            else{
                mis= mis+conf[i,j]
            }
        }
    }
    mis/(mis+clas)
}
```


```{r}
set.seed(1)
tr <- data.frame(matrix(ncol = length(names(data)), nrow = 0))
colnames(tr) <- names(data)
te<- data.frame(matrix(ncol = length(names(data)), nrow = 0))
colnames(te) <- names(data)
kerr <- data.frame(matrix(ncol = 1, nrow = 0))
kfold= 5
totsize=21
te.size= totsize/kfold

for (k in (1:50)){
    mis=0
    for (i in (1:kfold)){
        tr <- NULL
        te <- NULL
        te.idx= sample((1:totsize),te.size, replace=F)
        for (j in (1: totsize)){
            if (j %in% te.idx){
                te <- rbind(te,data[(data$subject== j),])
            }
            else{
                tr <- rbind(tr,data[(data$subject== j),])
            }
        }
        tr.x= tr[,-c(562:563)]
        tr.y= tr[,563]
        te.x= te[,-c(562:563)]
        te.y= te[,563]
        knn.testpred=knn(tr.x,te.x,tr.y,k=k)
        conf= table(knn.testpred,te.y)
        mis= mis + compute.misclass(conf)
    }
    mis=mis/kfold
    kerr <- rbind(kerr,mis)
}
```


```{r}
plot(c(1:50),kerr[,1],type='l',xlab='Number of Neighbours',ylab='Misclassification Rate')
```


```{r}
traindata <- data.frame(matrix(ncol = length(names(data)), nrow = 0))
testdata <- data.frame(matrix(ncol = length(names(data)), nrow = 0))
for (i in 1:nrow(data)){
  if(data$subject[i]<22){
    traindata= rbind(traindata,data[i,])
  }
  else{
    testdata= rbind(testdata,data[i,])
  }
}
```


```{r}
train.x= traindata[,-c(562,563)]
train.y= traindata[,563]
test.x= testdata[,-c(562,563)]
test.y= testdata[,563]
```


```{r}
knn.testpred=knn(train.x,test.x,train.y,k=17)
knn.testconf= table(knn.testpred,test.y)
knn.testconf
```


```{r}
accuracy= 1- compute.misclass(knn.testconf)
cat('The Accuracy of this model is: ', accuracy)
```


```{r}
cor = findCorrelation(cor(data[,-c(562,563)]), cutoff = 0.8)
uncor.data= data[,-cor]
```


```{r}
set.seed(1)
tr <- data.frame(matrix(ncol = length(names(uncor.data)), nrow = 0))
colnames(tr) <- names(uncor.data)
te<- data.frame(matrix(ncol = length(names(uncor.data)), nrow = 0))
colnames(te) <- names(uncor.data)
kerr <- data.frame(matrix(ncol = 1, nrow = 0))
kfold= 5
totsize=21
te.size= totsize/kfold

for (k in (1:50)){
    mis=0
    for (i in (1:kfold)){
        tr <- NULL
        te <- NULL
        te.idx= sample((1:totsize),te.size, replace=F)
        for (j in (1: totsize)){
            if (j %in% te.idx){
                te <- rbind(te,uncor.data[(uncor.data$subject== j),])
            }
            else{
                tr <- rbind(tr,uncor.data[(uncor.data$subject== j),])
            }
        }
        tr.x= tr[,-c(173:174)]
        tr.y= tr[,174]
        te.x= te[,-c(173:174)]
        te.y= te[,174]
        knn.testpred=knn(tr.x,te.x,tr.y,k=k)
        conf= table(knn.testpred,te.y)
        mis= mis + compute.misclass(conf)
    }
    mis=mis/kfold
    kerr <- rbind(kerr,mis)
}
```


```{r}
plot(c(1:50),kerr[,1],type='l',xlab='Number of Neighbours',ylab='Misclassification Rate')
```


```{r}
train.x= train.x[,-cor]
test.x= test.x[-cor]
```


```{r}
knn.testpred=knn(train.x,test.x,train.y,k=12)
knn.testconf= table(knn.testpred,test.y)
knn.testconf
```


```{r}
accuracy= 1- compute.misclass(knn.testconf)
cat('The Accuracy of this model is: ', accuracy)
```