Clustering Mango Dataset using K-Means
Muhammad Athanabil Andi Fawazdhia
2022-12-16
Preliminary
K-Means is one of many algorithms that can do clustering or unsupervised task. K-Means find label using metrics distances, widely using Euclidean distances. In this projects, i want to find label (large/medium/small) from Mango dataset using k-means
Library Used
library(factoextra)
library(tidyverse)
library(cluster)
library(ggplot2)
library(dplyr)
library(broom)
library(ggdendro)
library(readxl)
library(fmsb)
library(NbClust)
library(RColorBrewer)
library(gridExtra)Data Used
data <- read_excel("C:/Users/ASUS/Downloads/Nabil/Projek/Dataset_Mango Classification.xlsx")
data <- as.data.frame(data)
rownames(data) <- data$Mango_Types
data <- data[,-1:-2]
head(data)## Mango_Shape Mango_LeafShape Mango_Color Mango_Weight
## Madu Anggur Mango Round Crinkled Green 342.5
## Gedong Mango Round Flat Yellow 352.1
## Kemang Mango Oval Crinkled Green 571.9
## Chokanan Mango Kidney Shaped Flat Yellow 450.8
## Garifta Mango Kidney Shaped Flat Red 445.6
## Pakel Mango Round Flat Green 348.5
## Mango_Length Mango_Circumference
## Madu Anggur Mango 10.0 24.3
## Gedong Mango 11.5 24.4
## Kemang Mango 15.4 29.9
## Chokanan Mango 14.0 25.1
## Garifta Mango 13.5 26.8
## Pakel Mango 11.7 23.3Exploratory Data Analyst
Check Dimension Data
dim(data)## [1] 15 6Data have 15 rows and 6 columns
Check Type Each Variable
str(data)## 'data.frame': 15 obs. of 6 variables:
## $ Mango_Shape : chr "Round" "Round" "Oval" "Kidney Shaped" ...
## $ Mango_LeafShape : chr "Crinkled" "Flat" "Crinkled" "Flat" ...
## $ Mango_Color : chr "Green" "Yellow" "Green" "Yellow" ...
## $ Mango_Weight : num 342 352 572 451 446 ...
## $ Mango_Length : num 10 11.5 15.4 14 13.5 11.7 12.6 14.3 12.3 14.7 ...
## $ Mango_Circumference: num 24.3 24.4 29.9 25.1 26.8 23.3 24.8 27.6 25.8 27 ...Type variable Mango_Sahpe, Mango_LeafShape and Mango_Color as characther and the rest are numeric
Making Histogram for Each Numeric Variabel
par(mfrow=c(2,2))
hist(data[,4], col="yellow", main = "Mango_Weight")
hist(data[,5], col="red", main = "Mango_Length")
hist(data[,6], col="cyan", main = "Mango_Circumference")
Making Plot for Numeric Variable
plot(data, col="cyan")
Making Boxplot for Numeric Variable
boxplot(data[,4:6])
There are one outlier in Mango_Circumference
Preprocessing
Label Encoding
data$Mango_Shape <- as.numeric(factor(data$Mango_Shape))
data$Mango_LeafShape <- as.numeric(factor(data$Mango_LeafShape))
data$Mango_Color <- as.numeric(factor(data$Mango_Color))
head(data)## Mango_Shape Mango_LeafShape Mango_Color Mango_Weight
## Madu Anggur Mango 3 1 1 342.5
## Gedong Mango 3 2 3 352.1
## Kemang Mango 2 1 1 571.9
## Chokanan Mango 1 2 3 450.8
## Garifta Mango 1 2 2 445.6
## Pakel Mango 3 2 1 348.5
## Mango_Length Mango_Circumference
## Madu Anggur Mango 10.0 24.3
## Gedong Mango 11.5 24.4
## Kemang Mango 15.4 29.9
## Chokanan Mango 14.0 25.1
## Garifta Mango 13.5 26.8
## Pakel Mango 11.7 23.3Label Encoding have a goal to transform categorical data to numeric.
Normalization
data_norm <- scale(data)
head(data_norm)## Mango_Shape Mango_LeafShape Mango_Color Mango_Weight
## Madu Anggur Mango 1.3592235 -1.1660918 -0.7245688 -1.0897888
## Gedong Mango 1.3592235 0.1793987 1.6906606 -0.9575039
## Kemang Mango 0.1599086 -1.1660918 -0.7245688 2.0712694
## Chokanan Mango -1.0394062 0.1793987 1.6906606 0.4025503
## Garifta Mango -1.0394062 0.1793987 0.4830459 0.3308960
## Pakel Mango 1.3592235 0.1793987 -0.7245688 -1.0071108
## Mango_Length Mango_Circumference
## Madu Anggur Mango -1.9229867 -0.7558234
## Gedong Mango -0.9073247 -0.7005193
## Kemang Mango 1.7333965 2.3412092
## Chokanan Mango 0.7854453 -0.3133902
## Garifta Mango 0.4468913 0.6267804
## Pakel Mango -0.7719031 -1.3088650scale() is a function to do Normalization in R Studio, the goal is to change the scala data for example 1000,10,100 to range -3 to 3 using Z Score formula.
K-Means Clustering
set.seed(123)
#making prediction label using kmeans
km_label = kmeans(data_norm, 3)Making a Plot Clustering
#making plot cluster
fviz_cluster(km_label, data = data)
Plot result that k-means succes making 3 clusters from dataset.
Making a New DataFrame
#making new dataframe
df <- cbind(data, km_label$cluster)
df_final <- data.frame(df)
df_final## Mango_Shape Mango_LeafShape Mango_Color Mango_Weight
## Madu Anggur Mango 3 1 1 342.5
## Gedong Mango 3 2 3 352.1
## Kemang Mango 2 1 1 571.9
## Chokanan Mango 1 2 3 450.8
## Garifta Mango 1 2 2 445.6
## Pakel Mango 3 2 1 348.5
## Kweni Mango 3 3 1 368.6
## Manalagi Mango 1 2 1 440.2
## Dodol Mango 2 2 1 402.2
## Lalijiwo Mango 2 1 1 507.4
## Malibu Mango 1 1 2 416.4
## Arumanis Mango 1 3 1 298.1
## Legong Mango 2 3 2 465.3
## Gandaria Mango 1 2 3 420.5
## Golek Mango 2 1 1 493.7
## Mango_Length Mango_Circumference km_label.cluster
## Madu Anggur Mango 10.0 24.3 1
## Gedong Mango 11.5 24.4 1
## Kemang Mango 15.4 29.9 2
## Chokanan Mango 14.0 25.1 3
## Garifta Mango 13.5 26.8 3
## Pakel Mango 11.7 23.3 1
## Kweni Mango 12.6 24.8 1
## Manalagi Mango 14.3 27.6 2
## Dodol Mango 12.3 25.8 1
## Lalijiwo Mango 14.7 27.0 2
## Malibu Mango 12.5 25.5 3
## Arumanis Mango 11.3 22.5 1
## Legong Mango 14.2 25.5 3
## Gandaria Mango 12.6 26.0 3
## Golek Mango 12.0 26.5 2Making polygon plot
new <- aggregate(df[,-ncol(df)], list(df[,ncol(df)]), mean)
new## Group.1 Mango_Shape Mango_LeafShape Mango_Color Mango_Weight Mango_Length
## 1 1 2.50 2.166667 1.333333 352.00 11.56667
## 2 2 1.75 1.250000 1.000000 503.30 14.10000
## 3 3 1.20 2.000000 2.400000 439.72 13.36000
## Mango_Circumference
## 1 24.18333
## 2 27.75000
## 3 25.78000# Set graphic colors
library(RColorBrewer)
coul <- brewer.pal(3, 'Dark2')
colors_border <- coul
library(scales)
colors_in <- alpha(coul, 0.3)
radarchart( new[,-1], axistype=0 , maxmin=F,
#custom polygon
pcol=colors_border, pfcol=colors_in, plwd=4, plty=1,
#custom the grid
cglcol="grey", cglty=1, axislabcol="black", cglwd=0.8,
#custom labels
vlcex=0.8
)
# Add a legend
legend(x=1.7, y=1, legend = new$Group.1, bty = "n", pch=20, col=colors_in, text.col = "grey", cex=1.2, pt.cex=3)
Conclusion
Members of Cluster 2 show dominant value in Weight, length & circumference so we can conclude that mango in cluster 2 have large size, then followed up by cluster 3 as medium size, and cluster 1 as small size.