Assignment1_vneilley.Rmd

---
title: "Assignment 1 - HIA 218"
author: "vneilley@"
date: "3/15/2021"
output: html_document
---

## Pre-Work

**Install Pre-Requsites if needed**
if(!require(rmarkdown)) install.packages("rmarkdown") 
if(!require(ISLR)) install.packages("ISLR")
if(!require(class)) install.packages("class")
if(!require(dplyr)) install.packages("dplyr")
if(!require(caret)) install.packages("caret")
if(!require(ggplot)) install.packages("ggplot2")
if(!require(sqldf)) install.packages("sqldf")


### Import Data

```{r}
library(readr)
R_assignment_1_hcv_data <- read_csv("new- hcv data.csv")
View(R_assignment_1_hcv_data)
hcv_data <- R_assignment_1_hcv_data
hcv_manipulated <- R_assignment_1_hcv_data
```

## Question 1

**Age:** One outlier, left in

```{r}
boxplot(hcv_manipulated$Age,
        ylab = "Age")
```

**ALB:** Outliers above 70 deleted

```{r}
boxplot(hcv_manipulated$ALB,
        ylab = "ALB")

hcv_manipulated <- hcv_manipulated[!hcv_manipulated$ALB>70,]
```

**ALP:** Outliers above 300 deleted

```{r}
boxplot(hcv_manipulated$ALP,
        ylab = "ALP")

hcv_manipulated <- hcv_manipulated[!hcv_manipulated$ALP>300,]
```

**ALT:** Outliers above 150 deleted

```{r}
boxplot(hcv_manipulated$ALT,
        ylab = "ALT")

hcv_manipulated <- hcv_manipulated[!hcv_manipulated$ALT>150,]
```

**AST:** Outliers above 520 deleted

```{r}
boxplot(hcv_manipulated$AST,
        ylab = "AST")

hcv_manipulated <- hcv_manipulated[!hcv_manipulated$AST>250,]
```

**BIL:** Outliers above 150 deleted - there were significant amounts of outliers, and could not remove all or would significantly limit the dataset

```{r}
boxplot(hcv_manipulated$BIL,
        ylab = "BIL")

hcv_manipulated <- hcv_manipulated[!hcv_manipulated$BIL>150,]
```

**CHEL** No data removed \
**CHOL:** No data removed

```{r}
boxplot(hcv_manipulated$CHE,
        ylab = "CHE")

boxplot(hcv_manipulated$CHOL,
        ylab = "CHOL")
```

**CREA:** Outliers above 300 deleted

```{r}
boxplot(hcv_manipulated$CREA,
        ylab = "CREA")

hcv_manipulated <- hcv_manipulated[!hcv_manipulated$CREA>300,]
```

**GGT:** Outliers above 300 deleted

```{r}
boxplot(hcv_manipulated$GGT,
        ylab = "GGT")

hcv_manipulated <- hcv_manipulated[!hcv_manipulated$GGT>300,]
```

**PROT:** No data removed

```{r}
boxplot(hcv_manipulated$PROT,
        ylab = "PROT")
```

## Question 2

```{r}
hcv_manipulated <- na.omit(hcv_manipulated)
table(hcv_manipulated$final_category)
```

After removing some outliers, there are 42 positive data points, thus needing 65 negative individuals for a 40/60 split 

```{r}
library(sqldf)
negative <- sqldf('SELECT * FROM hcv_manipulated WHERE final_category = 1')
positive <- sqldf('SELECT * FROM hcv_manipulated WHERE final_category = 2')
```

### Randomize negative sample

```{r}
library(dplyr)
negative <- sample_n(negative, 65)
```

Combine the positive (42 obs) and negative (65 obs) datasets to get total with 107 observations with a 60/40 split of negative/postive

```{r}
total <- rbind(positive, negative)
total <- na.omit(total)
```

### Split 70/30 for training

```{r}
train_data <- total %>% mutate(ref=seq_len(nrow(total))) %>% sample_frac(0.7)
test_data <- total[-train_data$ref,]
```

### Manipulate and normalize train_data

```{r}
train_data <- train_data[-1]
train_data <- train_data[-1]
train_data <- train_data[-3]
train_data <- train_data[-14]
str(train_data)
```

```{r}
normal_train <- train_data %>% mutate_at(1:13, funs((.-min(.))/max(.-min(.))))
```

### Manipulate and normalize test_data

```{r}
test_data <- test_data[-1]
test_data <- test_data[-1]
test_data <- test_data[-3]
str(test_data)
```

```{r}
normal_test <- test_data %>% mutate_at(1:13, funs((.-min(.))/max(.-min(.))))
```

### Split training data

Extract 1st column of train dataset because it will be used for prediction.

```{r}
label_train = normal_train$final_category
normal_train <- normal_train[-1]
```

Find the number of observations. 
```{r}
NROW(label_train)
```
Since 75, the square of 75 is 8.7 so k = 9

### Split testing data

Extract 1st column of train dataset because it will be used for prediction.

```{r}
label_test = normal_test$final_category
normal_test <- normal_test[-1]
```

## Question 3

### Train KNN 

```{r}
library(class)
knn9 <- knn(train=normal_train, test=normal_test, cl=label_train, k=9)
```

### Test KNN

```{r}
ACC9 <- 100 * sum(label_test == knn9)/NROW(label_test)
table(knn9 ,label_test)
```

```{r}
caret::confusionMatrix(table(knn9 ,label_test))
```

Accuracy of 78.12% is reasonable for a model with a small dataset. Low specificity may suggest a poor mdoel as it will wrongfully tag negative individuals as positive, thus, causing potentially negative mental side effects.