customerSegmentation.Rmd

---
title: "Customer Segmentation"
author: "Rohit Tolawat"
date: "10/30/2020"
output: pdf_document
---

```{r}
pacman::p_load(caret, data.table, ggplot2,gridExtra,ggpubr,MASS,stats,data.table,factoextra)
knitr::opts_chunk$set(echo = FALSE, fig.width=12, fig.height=6, fig.path = 'Figs/')
options(digits = 3)
```


```{r}
set.seed(20)
customerData.df <- read.csv("C:/Users/rohit/Desktop/Analytics-Project/Customer-Segmentation-using-K-Means/customerSegmentation.csv")
customerNames <- customerData.df$Customer.Name
customerData.df <- customerData.df[,-1]
customerData.unscaled <- customerData.df
customerData.scaled <- scale(customerData.df, center = TRUE, scale = TRUE)

profit.scaled <- customerData.scaled[,1]
sales.scaled <- customerData.scaled[,2]

customerData.dt <- data.table(profit = profit.scaled, sales = sales.scaled)
str(customerData.dt)
```

```{r}
sum(is.na(customerData.dt$Profit))
sum(is.na(customerData.dt$Sales))

```
```{r}
summary(customerData.dt)
```


```{r}
ggplot(customerData.unscaled, aes(x=customerData.unscaled$Profit,y=customerData.unscaled$Sales))+
  geom_point()+
  xlab("Profit in $")+
  ylab("Sales in $")+
  geom_vline(xintercept = 0, linetype = "dashed", color = "blue")
```

```{r}
#The intention is to minimize the within sum of squared euclidean distance. This is the elbow method.
fviz_nbclust(customerData.unscaled, FUNcluster = kmeans, method = "wss")+
  ggtitle("Finding the right K for unscaled features")
```

```{r}
customerData.cluster.unscaled <- kmeans(customerData.unscaled, centers = 4,iter.max = 10)
customerData.cluster.unscaled$centers
```
```{r}
fviz_cluster(customerData.cluster.unscaled, data = customerData.unscaled, geom = c("point"),show.clust.cent = TRUE, main = "Customer clusters with unscaled features")+
  geom_vline(xintercept = 0, linetype="dashed", color = "black", size=0.5)
```

```{r}
plot1 <- ggplot(customerData.unscaled, aes(y=customerData.unscaled$Sales))+
  geom_boxplot()+
  ylab("Variation in Sales in $")+
  ggtitle("High variation in revenues")

plot2 <- ggplot(customerData.unscaled, aes(y=customerData.unscaled$Profit))+
  geom_boxplot()+
  ylab("Variation in Profit in $")+
  ggtitle("~50% of customers are loss-making")

grid.arrange(plot1, plot2, ncol=2)
```
```{r}
plot3 <- ggplot(customerData.dt, aes(y=customerData.dt$sales))+
  geom_boxplot()+
  ylab("Standardized sales")+
  ggtitle("Scaled revenues")

plot4 <- ggplot(customerData.dt, aes(y=customerData.dt$profit))+
  geom_boxplot()+
  ylab("Standardized profits")+
  ggtitle("Scaled customers")

grid.arrange(plot3, plot4, ncol=2)
```



```{r}
ggplot(customerData.dt, aes(x=customerData.dt$profit,y=customerData.dt$sales))+
  geom_point()+
  xlab("Profit in $")+
  ylab("Sales in $")+
  geom_vline(xintercept = 0, linetype = "dashed", color = "blue", size=0.5)
```

```{r}
hist(customerData.unscaled$Profit)
hist(customerData.dt$profit)
hist(customerData.unscaled$Sales)
hist(customerData.dt$sales)
```



```{r}
#The intention is to minimize the within sum of squared euclidean distance. This is the elbow method.
fviz_nbclust(customerData.dt, FUNcluster = kmeans, method = "wss")+
  ggtitle("Finding the right K with scaled features")
```



```{r}
customerData.cluster <- kmeans(customerData.dt, centers = 5,iter.max = 10)
customerData.cluster$centers
```




```{r}
fviz_cluster(customerData.cluster, data = customerData.dt, geom = c("point"),show.clust.cent = TRUE, main="Customer clusters after scaling features")+
  geom_vline(xintercept = 0, linetype="dashed", color = "black")+
  xlab("Profit in $")+
  ylab("Sales in $")
```

```{r}

clusterSize.dt <- data.frame(number = 1:length(unique(customerData.cluster$cluster)),size = customerData.cluster$size)

ggplot(clusterSize.dt, aes(x=number, y=size))+
  geom_bar(stat = "identity")+
  ggtitle("Cluster 2 is where I would begin my improvement initiatives")+
  xlab("Cluster number")+
  ylab("Cluster size")

```
```{r}
customerData.dt$cluster <- customerData.cluster$cluster
customerData.dt$customerName <- customerNames
str(customerData.dt)
write.csv(customerData.dt, "C:/Users/rohit/Desktop/Analytics-Project/Customer-Segmentation-using-K-Means/outputSegment.csv", row.names = FALSE)
customerData.cluster$size
```