# 安装必要的包install.packages(c("ggplot2", "randomForest", "calibrate"))library(ggplot2) # 用于数据可视化library(randomForest) # 用于构建随机森林模型
数据获取与加载
# 数据集URLURL <- "https://archive.ics.uci.edu/ml/machine-learning-databases/heart-disease/processed.cleveland.data"# 读取数据data <- read.csv(URL, header = FALSE)# 命名列names(data) <- c("age", "sex", "cp", "trestbps", "chol", "fbs", "restecg","thalach", "exang", "oldpeak", "slope", "ca", "thal", "hd")# 查看前6行数据head(data)
数据清洗与预处理
缺失值处理
将缺失值?替换为NA。数据类型转换
将部分变量转换为因子类型。
# 替换缺失值data[data == "?"] <- NA# 将性别和其他列转换为因子data$sex <- factor(ifelse(data$sex == 0, "F", "M"))data$cp <- factor(data$cp)data$ca <- factor(data$ca)data$thal <- factor(data$thal)data$hd <- factor(ifelse(data$hd == 0, "healthy", "unhealthy"))# 检查数据结构str(data)
缺失值填补
# 设置随机种子set.seed(123)# 使用randomForest填补缺失值data_imputed <- rfImpute(hd ~ ., data = data, ntree = 20)# 查看插补后的数据head(data_imputed)
构建随机森林模型
randomForest函数构建模型。# 构建随机森林模型set.seed(123)model <- randomForest(hd ~ ., data = data_imputed, proximity = TRUE)# 查看模型摘要print(model)
输出示例
OOB误差率: 83.5%
混淆矩阵:
树数量的优化
# 提取错误率数据error_rate <- data.frame(Trees = 1:500,Error = model$err.rate[,1])# 绘制误差率图ggplot(error_rate, aes(x = Trees, y = Error)) +geom_line(color = "blue") +labs(title = "OOB误差率随树数量变化", x = "树的数量", y = "OOB误差率")
变量随机抽样的优化
mtry 值（每次分裂所考虑的变量数量），找到最佳参数。# 初始化向量oob_errors <- numeric(10)# 测试不同的mtry值for (i in 1:10) {temp_model <- randomForest(hd ~ ., data = data_imputed, mtry = i, ntree = 500)oob_errors[i] <- temp_model$err.rate[500, 1]}# 绘制mtry与误差关系data_mtry <- data.frame(mtry = 1:10, OOB_Error = oob_errors)ggplot(data_mtry, aes(x = mtry, y = OOB_Error)) +geom_line(color = "red") +labs(title = "不同mtry值的OOB误差率", x = "mtry值", y = "OOB误差率")
样本相似性可视化（MDS图）
# 计算距离矩阵dist_matrix <- as.dist(1 - model$proximity)# MDS转换mds <- cmdscale(dist_matrix, eig = TRUE, k = 2)# 整理数据mds_data <- data.frame(X = mds$points[,1], Y = mds$points[,2], Class = data_imputed$hd)# 绘制MDS图ggplot(mds_data, aes(x = X, y = Y, color = Class)) +geom_point() +labs(title = "MDS多维尺度图", x = "第一维", y = "第二维")
