Perkembangan teknologi informasi memungkinkan pemanfaatan data untuk mengelompokkan objek secara efisien, termasuk pada biji kacang yang memiliki beragam karakteristik morfologis. Pengelompokan manual dinilai kurang efektif karena memakan waktu dan rentan terhadap subjektivitas, sehingga diperlukan pendekatan otomatis berbasis data. Clustering sebagai metode unsupervised learning dapat digunakan untuk mengelompokkan data berdasarkan kemiripan, dengan berbagai algoritma seperti K-Means, K-Median, DBSCAN, Mean Shift, dan Fuzzy C-Means yang memiliki keunggulan masing-masing. Untuk menentukan metode terbaik, dilakukan evaluasi menggunakan metrik seperti Silhouette Score dan Dunn Index. Selanjutnya, Exploratory Data Analysis (EDA) digunakan untuk memahami karakteristik tiap cluster sehingga menghasilkan insight yang lebih bermakna.

Import Library

library(tidyverse)
library(readxl)
library(flexclust)
library(dbscan)
library(meanShiftR)
library(e1071)
library(cluster)
library(fpc)

Load Dataset

data <- read_excel("D:/UAS tableau/Dry Bean Dataset.xlsx")
str(data)
## tibble [13,611 × 17] (S3: tbl_df/tbl/data.frame)
##  $ Area           : num [1:13611] 28395 28734 29380 30008 30140 ...
##  $ Perimeter      : num [1:13611] 610 638 624 646 620 ...
##  $ MajorAxisLength: num [1:13611] 208 201 213 211 202 ...
##  $ MinorAxisLength: num [1:13611] 174 183 176 183 190 ...
##  $ AspectRation   : num [1:13611] 1.2 1.1 1.21 1.15 1.06 ...
##  $ Eccentricity   : num [1:13611] 0.55 0.412 0.563 0.499 0.334 ...
##  $ ConvexArea     : num [1:13611] 28715 29172 29690 30724 30417 ...
##  $ EquivDiameter  : num [1:13611] 190 191 193 195 196 ...
##  $ Extent         : num [1:13611] 0.764 0.784 0.778 0.783 0.773 ...
##  $ Solidity       : num [1:13611] 0.989 0.985 0.99 0.977 0.991 ...
##  $ roundness      : num [1:13611] 0.958 0.887 0.948 0.904 0.985 ...
##  $ Compactness    : num [1:13611] 0.913 0.954 0.909 0.928 0.971 ...
##  $ ShapeFactor1   : num [1:13611] 0.00733 0.00698 0.00724 0.00702 0.0067 ...
##  $ ShapeFactor2   : num [1:13611] 0.00315 0.00356 0.00305 0.00321 0.00366 ...
##  $ ShapeFactor3   : num [1:13611] 0.834 0.91 0.826 0.862 0.942 ...
##  $ ShapeFactor4   : num [1:13611] 0.999 0.998 0.999 0.994 0.999 ...
##  $ Class          : chr [1:13611] "SEKER" "SEKER" "SEKER" "SEKER" ...
summary(data)
##       Area          Perimeter      MajorAxisLength MinorAxisLength
##  Min.   : 20420   Min.   : 524.7   Min.   :183.6   Min.   :122.5  
##  1st Qu.: 36328   1st Qu.: 703.5   1st Qu.:253.3   1st Qu.:175.8  
##  Median : 44652   Median : 794.9   Median :296.9   Median :192.4  
##  Mean   : 53048   Mean   : 855.3   Mean   :320.1   Mean   :202.3  
##  3rd Qu.: 61332   3rd Qu.: 977.2   3rd Qu.:376.5   3rd Qu.:217.0  
##  Max.   :254616   Max.   :1985.4   Max.   :738.9   Max.   :460.2  
##   AspectRation    Eccentricity      ConvexArea     EquivDiameter  
##  Min.   :1.025   Min.   :0.2190   Min.   : 20684   Min.   :161.2  
##  1st Qu.:1.432   1st Qu.:0.7159   1st Qu.: 36715   1st Qu.:215.1  
##  Median :1.551   Median :0.7644   Median : 45178   Median :238.4  
##  Mean   :1.583   Mean   :0.7509   Mean   : 53768   Mean   :253.1  
##  3rd Qu.:1.707   3rd Qu.:0.8105   3rd Qu.: 62294   3rd Qu.:279.4  
##  Max.   :2.430   Max.   :0.9114   Max.   :263261   Max.   :569.4  
##      Extent          Solidity        roundness       Compactness    
##  Min.   :0.5553   Min.   :0.9192   Min.   :0.4896   Min.   :0.6406  
##  1st Qu.:0.7186   1st Qu.:0.9857   1st Qu.:0.8321   1st Qu.:0.7625  
##  Median :0.7599   Median :0.9883   Median :0.8832   Median :0.8013  
##  Mean   :0.7497   Mean   :0.9871   Mean   :0.8733   Mean   :0.7999  
##  3rd Qu.:0.7869   3rd Qu.:0.9900   3rd Qu.:0.9169   3rd Qu.:0.8343  
##  Max.   :0.8662   Max.   :0.9947   Max.   :0.9907   Max.   :0.9873  
##   ShapeFactor1       ShapeFactor2        ShapeFactor3     ShapeFactor4   
##  Min.   :0.002778   Min.   :0.0005642   Min.   :0.4103   Min.   :0.9477  
##  1st Qu.:0.005900   1st Qu.:0.0011535   1st Qu.:0.5814   1st Qu.:0.9937  
##  Median :0.006645   Median :0.0016935   Median :0.6420   Median :0.9964  
##  Mean   :0.006564   Mean   :0.0017159   Mean   :0.6436   Mean   :0.9951  
##  3rd Qu.:0.007271   3rd Qu.:0.0021703   3rd Qu.:0.6960   3rd Qu.:0.9979  
##  Max.   :0.010451   Max.   :0.0036650   Max.   :0.9748   Max.   :0.9997  
##     Class          
##  Length:13611      
##  Class :character  
##  Mode  :character  
##                    
##                    
##

Preprocessing

label <- data$Class

df <- data %>% select(-Class)

df_scaled <- scale(df)
head(df_scaled)
##            Area  Perimeter MajorAxisLength MinorAxisLength AspectRation
## [1,] -0.8407176 -1.1432769       -1.306550      -0.6311299    -1.564995
## [2,] -0.8291572 -1.0138866       -1.395860      -0.4344286    -1.969712
## [3,] -0.8071275 -1.0787894       -1.252311      -0.5857131    -1.514236
## [4,] -0.7857117 -0.9771793       -1.278778      -0.4392741    -1.741554
## [5,] -0.7812103 -1.0973438       -1.380420      -0.2666536    -2.117915
## [6,] -0.7764701 -1.0283110       -1.255410      -0.4616520    -1.670900
##      Eccentricity ConvexArea EquivDiameter    Extent   Solidity roundness
## [1,]    -2.185640 -0.8414197    -1.0633015 0.2890768  0.3675999 1.4238148
## [2,]    -3.685904 -0.8260712    -1.0441784 0.6974512 -0.4628896 0.2310455
## [3,]    -2.045261 -0.8086740    -1.0080470 0.5781740  0.5183978 1.2528189
## [4,]    -2.742110 -0.7739468    -0.9733011 0.6712350 -2.2416847 0.5150303
## [5,]    -4.534862 -0.7842575    -0.9660443 0.4760028  0.8047429 1.8749235
## [6,]    -2.505324 -0.7781114    -0.9584197 0.5287761  0.5078901 1.1856533
##      Compactness ShapeFactor1 ShapeFactor2 ShapeFactor3 ShapeFactor4
## [1,]    1.839049    0.6807614     2.402085     1.925653    0.8383402
## [2,]    2.495358    0.3679534     3.100779     2.689603    0.7711101
## [3,]    1.764778    0.6031067     2.235009     1.841288    0.9167215
## [4,]    2.081639    0.4017030     2.514982     2.204169   -0.1979782
## [5,]    2.765229    0.1182639     3.270862     3.013352    0.9396054
## [6,]    2.007054    0.4046609     2.411297     2.118057    0.9555730

Clustering

set.seed(123)

# KMeans
kmeans_res <- kmeans(df_scaled, centers = 7, nstart = 25)
df_eval <- df_scaled[sample(1:nrow(df_scaled), 500), ]
kmeans_eval <- kmeans(df_eval, centers = 7)

# KMedian
kmedian_eval <- flexclust::kcca(df_eval, k = 7, family = kccaFamily("kmedians"))

# DBSCAN
dbscan_eval <- dbscan(df_eval, eps = 0.8, MinPts = 5)

# MeanShift
meanshift_eval <- meanShift(as.matrix(df_eval))

# FCM
fcm_eval <- cmeans(df_eval, centers = 7)

Evaluasi Metrik

evaluate_cluster <- function(data, cluster_result){
  sil <- silhouette(cluster_result, dist(data))
  sil_score <- mean(sil[,3])
  
  dunn <- cluster.stats(dist(data), cluster_result)$dunn
  
  return(c(silhouette = sil_score, dunn = dunn))
}

Hitung Evaluasi Metrik

# KMeans
eval_kmeans <- evaluate_cluster(df_eval, kmeans_eval$cluster)

# KMedian
eval_kmedian <- evaluate_cluster(df_eval, clusters(kmedian_eval))

# DBSCAN
db_cluster <- dbscan_eval$cluster
db_cluster[db_cluster == 0] <- NA

eval_dbscan <- evaluate_cluster(
  df_eval[!is.na(db_cluster),],
  db_cluster[!is.na(db_cluster)]
)

# MeanShift
eval_meanshift <- evaluate_cluster(df_eval, meanshift_eval$assignment)

# FCM
eval_fcm <- evaluate_cluster(df_eval, fcm_eval$cluster)

Hasil Evaluasi

results <- data.frame(
  Method = c("KMeans","KMedian","DBSCAN","MeanShift","FCM"),
  Silhouette = c(eval_kmeans[1],
                 eval_kmedian[1],
                 eval_dbscan[1],
                 eval_meanshift[1],
                 eval_fcm[1]),
  Dunn = c(eval_kmeans[2],
           eval_kmedian[2],
           eval_dbscan[2],
           eval_meanshift[2],
           eval_fcm[2])
)

print(results)
##      Method  Silhouette       Dunn
## 1    KMeans  0.28067439 0.05208664
## 2   KMedian  0.26942319 0.04399881
## 3    DBSCAN  0.16236248 0.15919198
## 4 MeanShift -0.06771737 0.06588645
## 5       FCM  0.26789887 0.03192598

Metode Terbaik

best_method <- results[which.max(results$Silhouette), ]
print(best_method)
##   Method Silhouette       Dunn
## 1 KMeans  0.2806744 0.05208664

EDA K-Means

data$Cluster <- kmeans_res$cluster

table(data$Cluster)
## 
##    1    2    3    4    5    6    7 
##  540 2481 3157 1769 3112  520 2032
cluster_summary <- data %>%
  group_by(Cluster) %>%
  summarise_all(mean)

print(cluster_summary)
## # A tibble: 7 × 18
##   Cluster    Area Perimeter MajorAxisLength MinorAxisLength AspectRation
##     <int>   <dbl>     <dbl>           <dbl>           <dbl>        <dbl>
## 1       1  65216.     1004.            386.            217.         1.79
## 2       2  74024.     1061.            394.            240.         1.65
## 3       3  31541.      658.            243.            165.         1.48
## 4       4  53677.      920.            374.            183.         2.04
## 5       5  44437.      794.            298.            190.         1.57
## 6       6 173668.     1586.            594.            375.         1.59
## 7       7  39393.      722.            249.            201.         1.24
## # ℹ 12 more variables: Eccentricity <dbl>, ConvexArea <dbl>,
## #   EquivDiameter <dbl>, Extent <dbl>, Solidity <dbl>, roundness <dbl>,
## #   Compactness <dbl>, ShapeFactor1 <dbl>, ShapeFactor2 <dbl>,
## #   ShapeFactor3 <dbl>, ShapeFactor4 <dbl>, Class <dbl>

Visualisasi

library(ggplot2)

pca <- prcomp(df_scaled)

pca_data <- data.frame(pca$x[,1:2])
pca_data$Cluster <- as.factor(data$Cluster)

ggplot(pca_data, aes(PC1, PC2, color = Cluster)) +
  geom_point() +
  ggtitle("Cluster KMeans")

ggplot(data, aes(x = as.factor(Cluster))) +
  geom_bar(fill = "skyblue") +
  ggtitle("Distribusi Cluster")

ggplot(data, aes(x = as.factor(Cluster), y = Area)) +
  geom_boxplot() +
  ggtitle("Area per Cluster")

ggplot(data, aes(x = as.factor(Cluster), y = Perimeter)) +
  geom_boxplot() +
  ggtitle("Perimeter per Cluster")