############################################################
# Identifying Institutional Peers through Cluster Analysis
# TAIR 2020 Conference
# Sample Code
# March 2, 2020
############################################################

#### Install packages used in code

install.packages(c("DMwR", "factoextra", "ggplot2", "gridExtra"))

##### Import data

peer_data <- read.csv("peer data.csv")


###############################
##### Step 1: Prepare Data ####
###############################

### 1A: Remove or estimate missing data

peer_data <- na.omit(peer_data) # removes all rows with missing values.


## or impute data using knnImputation from DMwR package

library(DMwR)

peer_data <- knnImputation(peer_data)


### 1B: Standardize data

scaled <- scale(peer_data) # scale() will only standardize numeric data.

####################################
#### Step 2: Calculate distance ####
####################################

library(factoextra)

distance <- get_dist(sample25) # computes distance matrix

# sample25 is a scaled sample of 25 institutions/rows used for the conference presentation.
# You can replace "sample25" with your scaled peer data.
# If you want other distance measures, change method parameter to "manhattan", "pearson", "spearman", or "kendall".
# Default method is "euclidean".
# Example: distanceMH <- get_dist(sample25, method="manhattan")

## plot distance matrix

fviz_dist(distance, gradient=list(low="#00AFBB", mid="white", high="#FC4E07"))

## use as.matrix() if you want to see distance values

distance_matrix <- as.matrix(distance)



##################################
#### Step 3: Cluster analysis ####
##################################

k2 <- kmeans(scaled, centers=2, nstart = 25) # performs clustering on matrix

fviz_cluster(k2, geom="point", data=scaled) # plots clusters

## add more clusters

k3 <- kmeans(scaled, centers = 3, nstart = 25)
k4 <- kmeans(scaled, centers = 4, nstart = 25)
k5 <- kmeans(scaled, centers = 5, nstart = 25)

## plots to compare clusters

p1 <- fviz_cluster(k2, geom = "point", data = scaled) + ggtitle("k = 2") +
  annotate("text", x=-1.25, y=-3.15, label="UH", size=3)

p2 <- fviz_cluster(k3, geom = "point",  data = scaled) + ggtitle("k = 3") +
  annotate("text", x=-1.25, y=-3.15, label="UH", size=3)

p3 <- fviz_cluster(k4, geom = "point",  data = scaled) + ggtitle("k = 4") +
  annotate("text", x=-1.25, y=-3.15, label="UH", size=3)

p4 <- fviz_cluster(k5, geom = "point",  data = scaled) + ggtitle("k = 5") +
  annotate("text", x=-1.25, y=-3.15, label="UH", size=3)

# Annotate is used in conference presentation. You may remove or add your own institution coordinates for labeling.

library(gridExtra)

grid.arrange(p1, p2, p3, p4, nrow = 2) # arranges clusters onto one page



#### choosing optimal clusters

set.seed(123) 
# random number generator used for
# simulating, generating random start points
# for reproducible results

## elbow method

fviz_nbclust(scaled, kmeans, method="wss")

## average silhouette

fviz_nbclust(scaled, kmeans, method="silhouette")


#### extract cluster results

print(k2)


###########################################################################################
# For  more information visit: https://jxmartinez.com/talk/identifying-institutional-peers/
###########################################################################################