Lab3.Rmd

---
output:
  pdf_document: default
  html_document: default
---
```{r, include=FALSE}
if (!require("tidyverse")) {
  install.packages("tidyverse")
  library(tidyverse)
}
if (!require("caret")) {
  install.packages("caret")
  library(caret)
}
```

```{r}
###################
##### Abalone #####
###################

# read dataset
abalone <- read.csv("~/DataStore-DataAnalytics/abalone_dataset.csv")

dataset <- abalone

## add new column age.group with 3 values based on the number of rings 
dataset$age.group <- cut(dataset$rings, br=c(0,8,11,35), labels = c("young", 'adult', 'old'))


## alternative way of setting age.group
dataset$age.group[dataset$rings<=8] <- "young"
dataset$age.group[dataset$rings>8 & dataset$rings<=11] <- "adult"
dataset$age.group[dataset$rings>11 & dataset$rings<=35] <- "old"

dataset <- na.omit(dataset)
```

```{r}
# Create dummy variables
dummies <- dummyVars(~ sex, data = dataset)

# Apply transformation
sex_dummies <- predict(dummies, newdata = dataset)

# Convert to dataframe and bind with original dataset (excluding 'sex' column)
dataset <- cbind(dataset[, !names(dataset) %in% "sex"], as.data.frame(sex_dummies))

dataset <- dataset %>% select(-rings, -sexM)

subset1 <- dataset[,1:8]
subset2 <- cbind(dataset[,9:10], dataset[,2:8])
```

```{r}
# sample create list of 105 (70% of 150) numbers randomly sampled from 1-150
set.seed(123)
n = 150
s1 <- sample(n,n*.8)
s2 <- sample(n,n*.8)

## create train & test sets based on sampled indexes 
s1train <- subset1[s1,]
s1test <- subset1[-s1,]

s2train <- subset2[s2,]
s2test <- subset2[-s2,]
```

# kNN Models:

```{r}
## train and evaluate multiple knn models to find optimal k
knn.model <- train(s1train[,1:7], s1train$age.group, method = "knn", tuneLength = 10, trControl = trainControl(method = "cv"))

# print model outputs
print(knn.model)

knn.predicted <- predict(knn.model, newdata = s1test[,1:7])

# create contingency table/ confusion matrix 
contingency.table <- table(knn.predicted, s1test$age.group, dnn=list('predicted','actual'))

contingency.table

# calculate classification accuracy
sum(diag(contingency.table))/length(s1test$age.group)
```
```{r}
## train and evaluate multiple knn models to find optimal k
knn.model2 <- train(s2train[,1:8], s2train$age.group, method = "knn", tuneLength = 10, trControl = trainControl(method = "cv"))

# print model outputs
print(knn.model2)

knn.predicted2 <- predict(knn.model2, newdata = s2test[,1:8])

# create contingency table/ confusion matrix 
contingency.table2 <- table(knn.predicted2, s2test$age.group, dnn=list('predicted','actual'))

contingency.table2

# calculate classification accuracy
sum(diag(contingency.table2))/length(s2test$age.group)
```
# K-Means Clustering
Subset 1 had the better results, so we proceed with doing K-means clustering on that dataset
```{r}
## run tests with multiple k values and plot WCSS
k.list <- c(2,3,4,5,6,7,8)

wcss.list <- c()

for (k in k.list) {
  
  km <- kmeans(subset1[,1:7], centers = k)
  
  wcss <- km$tot.withinss
    
  wcss.list <- c(wcss.list,wcss)
  
  ## get and plot clustering output 
  assigned.clusters <- as.factor(km$cluster)
  
  ggplot(subset1, aes(x = length, y = height, colour = assigned.clusters)) +
          geom_point()
  
}

plot(k.list,wcss.list,type = "b")


km <- kmeans(subset1[,1:7], centers = 4)
  
  wcss <- km$tot.withinss
    
  wcss.list <- c(wcss.list,wcss)
  
  ## get and plot clustering output 
  assigned.clusters <- as.factor(km$cluster)
  
  ggplot(subset1, aes(x = length, y = height, colour = assigned.clusters)) +
          geom_point()

```