Analisis Clustering Profil Nasabah Kredit Menggunakan MeanShift

1. Load dataset

Dataset yang digunakan berasal dari kaggle dengan link: https://www.kaggle.com/datasets/arjunbhasin2013/ccdata?utm_source=chatgpt.com

data <- read.csv("/Users/nabilanasywazen/Downloads/CC GENERAL.csv")
data

2. Seleksi Fitur

features <- data[, c('BALANCE', 
                     'CREDIT_LIMIT', 
                     'PAYMENTS', 
                     'MINIMUM_PAYMENTS', 
                     'PRC_FULL_PAYMENT', 
                     'CASH_ADVANCE', 
                     'CASH_ADVANCE_FREQUENCY', 
                     'TENURE')]
features

3. Handling Missing Value

3.1 Cek Missing Value Pada Fitur yang Dipilih

colSums(is.na(features))

               BALANCE           CREDIT_LIMIT               PAYMENTS       MINIMUM_PAYMENTS       PRC_FULL_PAYMENT           CASH_ADVANCE 
                     0                      1                      0                    313                      0                      0 
CASH_ADVANCE_FREQUENCY                 TENURE 
                     0                      0 

3.2 Drop Baris Data yang Mengandung Missing Value

df <- features[!is.na(features$MINIMUM_PAYMENTS) & !is.na(features$CREDIT_LIMIT), ]
colSums(is.na(df))

               BALANCE           CREDIT_LIMIT               PAYMENTS       MINIMUM_PAYMENTS       PRC_FULL_PAYMENT           CASH_ADVANCE 
                     0                      0                      0                      0                      0                      0 
CASH_ADVANCE_FREQUENCY                 TENURE 
                     0                      0 

4. Visualisasi Box Plot untuk Deteksi Outliers

library(ggplot2)
library(tidyr)
library(dplyr)

Attaching package: ‘dplyr’

The following objects are masked from ‘package:stats’:

    filter, lag

The following objects are masked from ‘package:base’:

    intersect, setdiff, setequal, union

numeric_data <- df %>% select_if(is.numeric)
long_data <- pivot_longer(numeric_data, cols = everything(), names_to = "Feature", values_to = "Value")

ggplot(long_data, aes(x = Value, y = Feature)) +
  geom_boxplot(fill = "lightblue", color = "navy") +
  theme_minimal() +
  labs(title = "Boxplot Semua Fitur", x = "Value", y = "Feature")

Boxplot menunjukkan hampir semua fitur yang digunakan mengandung outliers dalam jumlah yang signifikam. Terlihat jelas pada fitur “MINIMUM_PAYMENTS”, “PAYMENTS”, dan “CASH_ADVANCE” menunjukkan adanya variasi ekstrem dalam perilaku nasabah.

5. Normalisasi Data

Normalisasi pada semua fitur numerik dilakukan untuk membuat semua fitur berada pada range nilai yang sama.

minmax_scaled <- as.data.frame(lapply(df, function(x) (x - min(x)) / (max(x) - min(x))))
numeric_data <- minmax_scaled %>% select_if(is.numeric)
long_data <- pivot_longer(numeric_data, cols = everything(), names_to = "Feature", values_to = "Value")

ggplot(long_data, aes(x = Value, y = Feature)) +
  geom_boxplot(fill = "lightblue", color = "navy") +
  theme_minimal() +
  labs(title = "Boxplot Setelah Normalisasi", x = "Value", y = "Feature")

6. Clustering

Ketiga fitur ini dipilih karena melengkapi penggambaran menyeluruh tentang perilaku kredit nasabah.

• “BALANCE” = Menggambarkan tentang seberapa besar pinjaman credit nasabah
• “PAYMENTS” = Menggambarkan tentang pembayaram pinjaman credit yang telah dilakukan oleh nasabah
• “CREDIT_LIMIT” = Menggambarkan tentang seberapa bear kapasitas nasabah dalam memiliki pinjaman credit

library(meanShiftR)
library(scatterplot3d)

scaled_data <- as.matrix(minmax_scaled[, c("BALANCE", "PAYMENTS", "CREDIT_LIMIT")])
bandwidth_estimate <- rep(0.1293, ncol(scaled_data))

ms_result <- meanShift(scaled_data, bandwidth = bandwidth_estimate)
cluster_centers <- ms_result$centers

colors <- rainbow(length(unique(ms_result$assignment)))
plot3d <- scatterplot3d(scaled_data[, 1], scaled_data[, 2], scaled_data[, 3], 
                        pch = 16, 
                        color = colors[ms_result$assignment],
                        xlab = colnames(scaled_data)[1], 
                        ylab = colnames(scaled_data)[2], 
                        zlab = colnames(scaled_data)[3],
                        main = "MeanShift Clustering 3D Plot")

plot3d$points3d(cluster_centers[, 1], cluster_centers[, 2], cluster_centers[, 3],
                col = "black", pch = 4, cex = 3, lwd = 2) 

Clustering menggunakan MeanSHift dengan bandwidth 0.1293 dihasilkan 14 cluster dengan visualisasi warna yang berbeda-beda.

length(unique(ms_result$assignment))

[1] 14

7. Analisis Karakteristik Tiap CLuster

7.1 Menghitung Rata-rata, Nilai Min, dan Nilai Max Untuk Tiap Fitur

library(dplyr)

df$Cluster <- as.factor(ms_result$assignment)
cluster_summary <- df %>%
  group_by(Cluster) %>%
  summarise(
    avg_balance = mean(BALANCE, na.rm = TRUE),
    avg_payments = mean(PAYMENTS, na.rm = TRUE),
    avg_credit_limit = mean(CREDIT_LIMIT, na.rm = TRUE),
    min_balance = min(BALANCE, na.rm = TRUE),
    max_balance = max(BALANCE, na.rm = TRUE),
    min_payments = min(PAYMENTS, na.rm = TRUE),
    max_payments = max(PAYMENTS, na.rm = TRUE),
    min_credit_limit = min(CREDIT_LIMIT, na.rm = TRUE),
    max_credit_limit = max(CREDIT_LIMIT, na.rm = TRUE)
  )

print(cluster_summary)

7.2 Analisis Tingkat Risiko Tiap Cluster

library(dplyr)

cluster_risk <- cluster_summary %>%
  mutate(
    bcl_ratio = avg_balance / avg_credit_limit,              
    pb_ratio  = avg_payments / avg_balance,                  
    risk_level = case_when(
      bcl_ratio >= 0.8 & pb_ratio < 0.5  ~ "Tinggi",
      bcl_ratio >= 0.6 & pb_ratio < 1.0  ~ "Sedang",
      pb_ratio >= 1.0                    ~ "Rendah",
      TRUE                               ~ "Sedang"  
    )
  )

# Lihat hasil akhir
print(cluster_risk %>% select(Cluster, avg_balance, avg_payments, avg_credit_limit, bcl_ratio, pb_ratio, risk_level))

8. Evaluasi Silhouette Score

Silhouette Index digunakan untuk menggambarkan seberapa baik kualitas hasil clustering yang telah dilakukan.

• Nilai mendekati 1 berarti data tersebut cocok dengan klaster data itu sendiri dan jauh dari kalster lain
• Nilai mendekati 0 berarti data mengalami overlapping
• Nilai mendekati -1 data mungkin mengalami salah clustering

library(cluster)

sil <- silhouette(ms_result$assignment, dist(scaled_data))
mean_silhouette_score <- mean(sil[, 3])
print(mean_silhouette_score)

[1] 0.7425532