BC_TumorAnalysis_Clustering.Rmd

---
title: "Tumor Analysis- Cluster Analysis"
description: "Develop accurate models for tumor diagnosis prediction using comprehensive analysis of Wisconsin Breast Cancer Diagnostic (WBCD) dataset including feature selection, classification models, visualization, outlier identification, imbalanced data management, performance evaluation, and generalization approaches. Increased model accuracy and precision with PCA for dimensionality reduction and K-means clustering for feature extraction"
author: "pk673@rutgets.edu"
date: "2023-04-07"
output: html_document
---

```{r}

library(readr)
library(MVA)
library(HSAUR2)
library(SciViews)
library(scatterplot3d)
library(car)
library(lattice)
library(GGally)
library(ggplot2)
library(ggridges)
library(ggvis)
library(ggthemes)
library(cowplot)
library(gapminder)
library(gganimate)
library(dplyr)
library(tidyverse)
library(grid)
library(gridExtra)
library(RColorBrewer)
library(Hotelling)
library(stats)
library(biotools)
library(factoextra)
library(FactoMineR)
library(ggfortify)
library(psych)
library(corrplot)
library(MASS)
library(ggplot2)
library(memisc)
library(ROCR)
library(dplyr)
library(klaR)
library(reshape2)
library(corrplot)
library(MASS)
library(ggplot2)
library(memisc)
library(ROCR)
library(dplyr)
library(klaR)
library(plyr)

# Load the WBCD dataset from the UCI repository
wbdc <- read.csv(url(
  "https://archive.ics.uci.edu/ml/machine-learning-databases/breast-cancer-wisconsin/wdbc.data"),
                     header = FALSE,
                     col.names = c("ID", "diagnosis", "radius_mean", 
                                   "texture_mean", "perimeter_mean", 
                                   "area_mean", "smoothness_mean", 
                                   "compactness_mean", "concavity_mean", 
                                   "concave_pts_mean", "symmetry_mean", 
                                   "fractal_dim_mean", "radius_se", 
                                   "texture_se", "perimeter_se", "area_se", 
                                   "smoothness_se", "compactness_se", 
                                   "concavity_se", "concave_pts_se", 
                                   "symmetry_se", "fractal_dim_se", 
                                   "radius_worst", "texture_worst", 
                                   "perimeter_worst", "area_worst", 
                                   "smoothness_worst", "compactness_worst", 
                                   "concavity_worst", "concave_pts_worst", 
                                   "symmetry_worst", "fractal_dim_worst"))
dim(wbdc)
diagnosis <- factor(wbdc$diagnosis)
str(wbdc)
# Data Preprocessing
rownames(wbdc)=wbdc$id
wbdc=wbdc[,-1]
wbdc=wbdc[,-1]
#fix(wbdc)

#inspection of missing values in the dataset
sum(is.na(wbdc)) # number of records with N/A values


set.seed(123)
#any(is.na(wbdc)) # Check if there are any missing values in the data
#any(!is.finite(wbdc)) # Check if there are any infinite values in the data
#wbdc_mat <- as.matrix(wbdc)
k_means_fit <- kmeans(wbdc, 2,nstart = 100) # k = 2
print(k_means_fit)
table(k_means_fit$cluster)

#cluster mean
aggregate(wbdc, by=list(cluster=k_means_fit$cluster), mean)

#visualisation
library(factoextra)
wbdc_num <- wbdc[,0:ncol(wbdc)]

# Convert to matrix
wbdc_mat <- as.matrix(wbdc_num)

# Cluster the data
k_means_fit <- kmeans(wbdc_mat, centers = 2, nstart = 25)

# Visualize the clusters
library(factoextra)
library(ggplot2)
fviz_cluster(k_means_fit, data = wbdc_mat, 
             geom = "point", 
             palette = c("#2E9FDF", "#E7B800"),
             ellipse.type = "euclid",
             star.plot = TRUE, 
             repel = FALSE, 
             ggtheme = theme_minimal()
)

# how many clusters for K-means 
wssplot <- function(wbdc_mat, nc=10, seed=1234){
  wss <- (nrow(wbdc_mat)-1)*sum(apply(wbdc_mat,2,var))
  for (i in 2:nc){
    set.seed(seed)
    wss[i] <- sum(kmeans(wbdc_mat, centers=i)$withinss)}
  plot(1:nc, wss, type="b", xlab="Number of Clusters",
       ylab="Within groups sum of squares")}

wssplot(wbdc_mat, nc=10) 

library(factoextra)
fviz_nbclust(wbdc_mat, kmeans, method = "wss") +
  geom_vline(xintercept = 2, linetype = 2)+
  labs(subtitle = "Elbow method")


fviz_cluster(kmeans(wbdc_mat, centers = 2), geom = "point", data = wbdc_mat)

# Elbow method
fviz_nbclust(wbdc, kmeans, method ="wss") +
  geom_vline(xintercept = 2, linetype = 2)+
  labs(subtitle = "Elbow method")
# Silhouette method
fviz_nbclust(wbdc, kmeans, method = "silhouette")+
  labs(subtitle = "Silhouette method")

#############################################
# hierarchical clustering 

#As part of the preparation for hierarchical clustering, the distance between all pairs
# of observations are computed. Furthermore, there are different ways to link clusters 
#together, with single, complete, and average being the most common linkage methods.


# Perform PCA
pca <- prcomp(wbdc, scale = TRUE)

# Extract scores
scores <- pca$x

# Compute Euclidean distance matrix
d <- dist(scores, method = "euclidean")
d_matrix <- as.matrix(d)
#d_matrix

#agglomerative clustering
d <- dist(scores, method = "euclidean") # Euclidean distance matrix.
d_matrix=as.matrix(d)
#d_matrix

#complete linkage
H.fit <- hclust(d, method="complete")
fviz_dend(H.fit, cex = 0.5)
# Average linkage
H.fit.avg=hclust(d, method="average")
fviz_dend(H.fit.avg)
# Ward linkage
H.fit.ward=hclust(d, method="ward.D2")
fviz_dend(H.fit.ward)

# Cut tree into 2/4 groups
grp <- cutree(H.fit.ward, k = 2)
head(grp, n = 2)
# Number of members in each cluster

table(grp,diagnosis)
table(k_means_fit$cluster, grp)

fviz_dend(H.fit.ward, k = 2, 
          cex = 0.5,
          k_colors = c("#2E9FDF", "yellow","red","pink"),
          color_labels_by_k = TRUE,
          rect = TRUE 
)
fviz_cluster(list(data = wbdc, cluster = grp),
             palette = c("#2E9FDF", "#E7B800", "#FC4E07"),
             ellipse.type = "convex", 
             repel = FALSE,
             show.clust.cent = FALSE, ggtheme = theme_minimal())
# # ##########################################################
# ## validation
# 
# km2 <- data.frame(wbdc, cluster = k_means_fit$cluster, diagnosis = wbdc$diagnosis)
# km2$diagnosis <- as.factor(mapvalues(km2$diagnosis, from=c("B", "M"), to=c("2", "1")))
# mean(km2$cluster == km2$diagnosis)
# # Output : 0.8541301
# hc2 <- data.frame(wbdc, grp, diagnosis = wbdc$diagnosis)
# hc2$diagnosis <- as.factor(mapvalues(hc2$diagnosis,
#                                       from = c("B", "M"),
#                                       to = c("2","1")))
# 
# mean(hc2$grp == hc2$diagnosis)
# #Output: 0.8804921


```