ANN Clasification

Menyiapkan library dan data

data <- read_xlsx("bank_latih_preprocessed.xlsx")
str(data)

## tibble [4,521 × 41] (S3: tbl_df/tbl/data.frame)
##  $ Age              : num [1:4521] 30 33 35 30 59 35 36 39 41 43 ...
##  $ job              : chr [1:4521] "unemployed" "services" "management" "management" ...
##  $ marital          : chr [1:4521] "married" "married" "single" "married" ...
##  $ education        : chr [1:4521] "primary" "secondary" "tertiary" "tertiary" ...
##  $ default          : chr [1:4521] "no" "no" "no" "no" ...
##  $ balance          : num [1:4521] 1787 4789 1350 1476 0 ...
##  $ housing          : chr [1:4521] "no" "yes" "yes" "yes" ...
##  $ loan             : chr [1:4521] "no" "yes" "no" "yes" ...
##  $ contact          : chr [1:4521] "cellular" "cellular" "cellular" "unknown" ...
##  $ day              : num [1:4521] 19 11 16 3 5 23 14 6 14 17 ...
##  $ month            : chr [1:4521] "oct" "may" "apr" "jun" ...
##  $ duration         : num [1:4521] 79 220 185 199 226 141 341 151 57 313 ...
##  $ campaign         : num [1:4521] 1 1 1 4 1 2 1 2 2 1 ...
##  $ pdays            : num [1:4521] -1 339 330 -1 -1 176 330 -1 -1 147 ...
##  $ previous         : num [1:4521] 0 4 1 0 0 3 2 0 0 2 ...
##  $ poutcome         : chr [1:4521] "unknown" "failure" "failure" "unknown" ...
##  $ y                : chr [1:4521] "no" "no" "no" "no" ...
##  $ Age_z            : num [1:4521] -1.056 -0.772 -0.583 -1.056 1.686 ...
##  $ balance_z        : num [1:4521] 0.1211 1.1185 -0.0241 0.0177 -0.4727 ...
##  $ day_z            : num [1:4521] 0.374 -0.596 0.0103 -1.5659 -1.3234 ...
##  $ duration_z       : num [1:4521] -0.712 -0.169 -0.304 -0.25 -0.146 ...
##  $ campaign_z       : num [1:4521] -0.577 -0.577 -0.577 0.388 -0.577 ...
##  $ pdays_z          : num [1:4521] -1 0.974 0.897 -1 -1 ...
##  $ previous_z       : num [1:4521] -0.32 2.04 0.27 -0.32 -0.32 ...
##  $ Age_norm         : num [1:4521] 0.162 0.206 0.235 0.162 0.588 ...
##  $ balance_norm     : num [1:4521] 0.0685 0.1088 0.0626 0.0643 0.0445 ...
##  $ day_norm         : num [1:4521] 0.6 0.3333 0.5 0.0667 0.1333 ...
##  $ duration_norm    : num [1:4521] 0.0248 0.0715 0.0599 0.0645 0.0735 ...
##  $ campaign_norm    : num [1:4521] 0 0 0 0.0612 0 ...
##  $ pdays_norm       : num [1:4521] -1 0.389 0.378 -1 -1 ...
##  $ previous_norm    : num [1:4521] 0 0.16 0.04 0 0 0.12 0.08 0 0 0.08 ...
##  $ job_encoded      : num [1:4521] 11 8 5 5 2 5 7 10 3 8 ...
##  $ marital_encoded  : num [1:4521] 2 2 3 2 2 3 2 2 2 2 ...
##  $ education_encoded: num [1:4521] 1 2 3 3 2 3 3 2 3 1 ...
##  $ default_encoded  : num [1:4521] 1 1 1 1 1 1 1 1 1 1 ...
##  $ housing_encoded  : num [1:4521] 1 2 2 2 2 1 2 2 2 2 ...
##  $ loan_encoded     : num [1:4521] 1 2 1 2 1 1 1 1 1 2 ...
##  $ contact_encoded  : num [1:4521] 1 1 1 3 3 1 1 1 3 1 ...
##  $ month_encoded    : num [1:4521] 11 9 1 7 9 4 9 9 9 1 ...
##  $ poutcome_encoded : num [1:4521] 4 1 1 4 4 1 2 4 4 1 ...
##  $ y_encoded        : num [1:4521] 1 1 1 1 1 1 1 1 1 1 ...

Mengubah variabel target menjadi bentuk biner

data$y_binary <- ifelse(data$y == "yes", 1, 0) 

Memilih fitur

features.1 <- c("Age", "balance", "day", "duration", "campaign", "pdays", "previous", "job_encoded", "marital_encoded", "education_encoded", "default_encoded", "housing_encoded", "loan_encoded", "contact_encoded", "month_encoded", "poutcome_encoded")

features.2_1 <- c("Age_z", "balance_z", "day", "duration_z", "campaign_z", "pdays_z", "previous_z", "job_encoded", "marital_encoded", "education_encoded", "default_encoded", "housing_encoded", "loan_encoded", "contact_encoded", "month_encoded", "poutcome_encoded")

features.3_1 <- c("Age_norm", "balance_norm", "day", "duration_norm", "campaign_norm", "pdays_norm", "previous_norm", "job_encoded", "marital_encoded", "education_encoded", "default_encoded", "housing_encoded", "loan_encoded", "contact_encoded", "month_encoded", "poutcome_encoded")

model.1_data <- data[, c(features.1, "y_binary")]
model.2_1_data <- data[, c(features.2_1, "y_binary")]
model.3_1_data <- data[, c(features.3_1, "y_binary")]

Split data training dan testing

# Split data
set.seed(123)
index <- createDataPartition(model.1_data$y_binary, p = 0.8, list = FALSE)
trainData_1 <- model.1_data[index, ]
testData_1 <- model.1_data[-index, ]

trainData_2_1 <- model.2_1_data[index, ]
testData_2_1 <- model.2_1_data[-index, ]

trainData_3_1 <- model.3_1_data[index, ]
testData_3_1 <- model.3_1_data[-index, ]

Formula untuk model training

# Train model
formula.1 <- as.formula(paste("y_binary ~", paste(names(trainData_1)[!names(trainData_1) %in% c("y_binary")], collapse = " + ")))

formula.2_1 <- as.formula(paste("y_binary ~", paste(names(trainData_2_1)[!names(trainData_2_1) %in% c("y_binary")], collapse = " + ")))

formula.3_1 <- as.formula(paste("y_binary ~", paste(names(trainData_3_1)[!names(trainData_3_1) %in% c("y_binary")], collapse = " + ")))

Model dengan fitur asli

ann_model.1<- neuralnet(formula.1, data = trainData_1, hidden = c(5), linear.output = FALSE)

Model dengan fitur hasil standardisasi (Z-Score)

ann_model.2_1 <- neuralnet(formula.2_1, data = trainData_2_1, hidden = c(5), linear.output = FALSE)

Model dengan fitur hasil normalisasi (min-max)

ann_model.3_1 <- neuralnet(formula.3_1, data = trainData_3_1, hidden = c(5), linear.output = FALSE)

Plot model ANN dengan fitur asli

plot(ann_model.1)

Plot model ANN dengan fitur hasil standardisasi

plot(ann_model.2_1)

Plot model ANN dengan fitur hasil normalisasi

plot(ann_model.3_1)

Metrik evaluasi setiap model

# Simpan semua model, data, dan fitur dalam list
models <- list(ann_model.1, ann_model.2_1, ann_model.3_1)
train_data <- list(trainData_1, trainData_2_1, trainData_3_1)
test_data <- list(testData_1, testData_2_1, testData_3_1)
features_list <- list(features.1, features.2_1, features.3_1)

# Fungsi bantu evaluasi
evaluate_model <- function(model, train_set, test_set, features, index) {
  cat("\n===============================\n")
  cat(paste("📊 Evaluasi Model ANN -", index, "\n"))
  cat("===============================\n")
  
  # ---- TRAINING ----
  pred_train <- predict(model, train_set[, features], type = "raw")
  pred_train_class <- ifelse(pred_train > 0.5, 1, 0)
  
  cm_train <- confusionMatrix(as.factor(pred_train_class), as.factor(train_set$y_binary))
  precision_train <- Precision(pred_train_class, train_set$y_binary)
  recall_train <- Recall(pred_train_class, train_set$y_binary)
  f1_train <- F1_Score(pred_train_class, train_set$y_binary)
  accuracy_train <- Accuracy(pred_train_class, train_set$y_binary)
  
  cat("\n📌 TRAINING SET:\n")
  print(cm_train$table)
  cat(sprintf("Accuracy : %.4f | Precision: %.4f | Recall: %.4f | F1: %.4f\n",
              accuracy_train, precision_train, recall_train, f1_train))
  
  # ---- TESTING ----
  pred_test <- predict(model, test_set[, features], type = "raw")
  pred_test_class <- ifelse(pred_test > 0.5, 1, 0)
  
  cm_test <- confusionMatrix(as.factor(pred_test_class), as.factor(test_set$y_binary))
  precision_test <- Precision(pred_test_class, test_set$y_binary)
  recall_test <- Recall(pred_test_class, test_set$y_binary)
  f1_test <- F1_Score(pred_test_class, test_set$y_binary)
  accuracy_test <- Accuracy(pred_test_class, test_set$y_binary)
  
  cat("\n📌 TESTING SET:\n")
  print(cm_test$table)
  cat(sprintf("Accuracy : %.4f | Precision: %.4f | Recall: %.4f | F1: %.4f\n",
              accuracy_test, precision_test, recall_test, f1_test))
}

# Loop untuk evaluasi semua model
for (i in 1:3) {
  evaluate_model(models[[i]], train_data[[i]], test_data[[i]], features_list[[i]], i)
}

## 
## ===============================
## 📊 Evaluasi Model ANN - 1 
## ===============================
## 
## 📌 TRAINING SET:
##           Reference
## Prediction    0    1
##          0 3190  394
##          1   13   20
## Accuracy : 0.8875 | Precision: 0.9959 | Recall: 0.8901 | F1: 0.9400
## 
## 📌 TESTING SET:
##           Reference
## Prediction   0   1
##          0 797 106
##          1   0   1
## Accuracy : 0.8827 | Precision: 1.0000 | Recall: 0.8826 | F1: 0.9376
## 
## ===============================
## 📊 Evaluasi Model ANN - 2 
## ===============================
## 
## 📌 TRAINING SET:
##           Reference
## Prediction    0    1
##          0 3103  188
##          1  100  226
## Accuracy : 0.9204 | Precision: 0.9688 | Recall: 0.9429 | F1: 0.9557
## 
## 📌 TESTING SET:
##           Reference
## Prediction   0   1
##          0 761  73
##          1  36  34
## Accuracy : 0.8794 | Precision: 0.9548 | Recall: 0.9125 | F1: 0.9332
## 
## ===============================
## 📊 Evaluasi Model ANN - 3 
## ===============================
## 
## 📌 TRAINING SET:
##           Reference
## Prediction    0    1
##          0 3115  195
##          1   88  219
## Accuracy : 0.9218 | Precision: 0.9725 | Recall: 0.9411 | F1: 0.9565
## 
## 📌 TESTING SET:
##           Reference
## Prediction   0   1
##          0 758  67
##          1  39  40
## Accuracy : 0.8827 | Precision: 0.9511 | Recall: 0.9188 | F1: 0.9346