使用基于xgbtree的REF进行特征选择，请写出详细正确无误的R语言代码

以下是使用基于xgbtree的REF进行特征选择的R语言代码：

# 导入数据
data <- read.csv("data.csv")

# 分割数据集为训练集和测试集
train_index <- sample(1:nrow(data), 0.7 * nrow(data))
train_data <- data[train_index, ]
test_data <- data[-train_index, ]

# 安装和载入xgboost包
install.packages("xgboost")
library(xgboost)

# 训练xgboost模型
xgb_model <- xgboost(data = as.matrix(train_data[, -ncol(train_data)]), 
                     label = train_data$label, 
                     objective = "binary:logistic",
                     nrounds = 100,
                     verbose = FALSE)

# 计算特征重要性排名
importance <- xgb.importance(colnames(train_data[, -ncol(train_data)]), 
                              model = xgb_model, 
                              data = train_data[, -ncol(train_data)], 
                              label = train_data$label,
                              type = "gain")
# 打印特征重要性排名
print(importance)

# 基于xgbtree的REF特征选择
selected_features <- colnames(train_data[, -ncol(train_data)])
for (i in 1:(ncol(train_data) - 1)) {
  # 移除最不重要的特征
  least_important_feature <- importance$Feature[which.min(importance$Gain)]
  selected_features <- selected_features[selected_features != least_important_feature]
  
  # 训练xgboost模型并计算测试集上的性能
  xgb_model <- xgboost(data = as.matrix(train_data[, selected_features]), 
                       label = train_data$label, 
                       objective = "binary:logistic",
                       nrounds = 100,
                       verbose = FALSE)
  pred <- predict(xgb_model, as.matrix(test_data[, selected_features]))
  auc <- caret::twoClassSummary(test_data$label, pred, metrics = "ROC")
  print(paste0("Selected features: ", paste(selected_features, collapse = ", ")))
  print(paste0("AUC: ", auc$ROC))
  
  # 如果性能下降，则停止特征移除
  if (i > 1 & auc$ROC <= prev_auc) {
    break
  }
  prev_auc <- auc$ROC
}

上述代码中，首先导入数据并将其分割为训练集和测试集。然后安装和载入xgboost包，并使用xgboost训练模型和计算特征重要性排名。接下来，使用基于xgbtree的REF进行特征选择，每次移除最不重要的特征，并重新训练模型并计算测试集上的性能。如果性能有所下降，则停止特征移除。最终得到的特征子集即为最佳特征子集。