一篇学会！目前最常用，最全面的机器学习模型代码分享~

# 安装必要的包（如果尚未安装）install.packages(c("caret", "randomForest", "glmnet", "e1071", "xgboost",                    "rpart", "kernlab", "nnet", "pROC", "ggplot2"))
# 加载包library(caret)      # 机器学习统一接口library(randomForest) # 随机森林library(glmnet)     # 正则化回归library(e1071)      # SVMlibrary(xgboost)    # XGBoostlibrary(rpart)      # 决策树library(kernlab)    # 其他核方法library(nnet)       # 神经网络library(pROC)       # ROC曲线library(ggplot2)    # 可视化
# 设置随机种子保证可重复性set.seed(123)
# 加载示例数据集（使用内置的iris数据集）


    
data(iris)

# 划分训练集和测试集trainIndex 0.7, list = FALSE)trainData testData 
# 标准化数据（对数值变量）preProcValues 5], method = c("center", "scale"))trainTransformed testTransformed 
# 查看数据结构str(trainTransformed)

# 训练逻辑回归模型logitModel ~ .,                  data = trainTransformed,                  family = binomial(link = "logit"))
# 多分类逻辑回归multinomModel ~ ., data = trainTransformed)
# 预测logitPred "response")multinomPred 
# 评估confusionMatrix(multinomPred, testTransformed$Species)

#正则化回归（lasso、ridge）# 准备矩阵格式的数据x ~ ., trainTransformed)[, -1]y $Species
# 交叉验证选择最佳lambdacvFit "multinomial", alpha = 1) # LASSO (alpha=1)
# 查看结果plot(cvFit)
# 预测ridgePred                     newx = model.matrix(Species ~ ., testTransformed)[, -1],                    s = "lambda.min",                    type = "class")
# 评估confusionMatrix(factor(ridgePred), testTransformed$Species)

# 训练决策树treeModel ~ ., data = trainTransformed, method = "class")
# 可视化决策树plot(treeModel)text(treeModel, use.n = TRUE)
# 预测treePred "class")
# 评估confusionMatrix(treePred, testTransformed$Species)

# 训练随机森林rfModel ~ ., data = trainTransformed, ntree = 500)
# 查看变量重要性varImpPlot(rfModel)
# 预测rfPred 
# 评估confusionMatrix(rfPred, testTransformed$Species)

# 训练SVM模型svmModel ~ ., data = trainTransformed, kernel = "radial")
# 预测svmPred 
# 评估confusionMatrix(svmPred, testTransformed$Species)
# 使用caret调参svmTune ~ .,                data = trainTransformed,                method = "svmRadial",                tuneLength = 9,                trControl = trainControl(method = "cv"))

# 准备数据trainX ~ ., trainTransformed)[, -1]trainY $Species) - 1testX ~ ., testTransformed)[, -1]
# 训练XGBoost模型xgbModel                    label = trainY,                   nrounds = 100,                   objective = "multi:softprob",                   num_class = length(levels(iris$Species)),                   eval_metric = "mlogloss")
# 预测xgbPred xgbPred $Species)[max.col(xgbPred)]
# 评估confusionMatrix(factor(xgbPred), testTransformed$Species)

# 训练神经网络nnModel                data = trainTransformed,               size = 5,               decay = 0.01,               maxit = 200,               trace = FALSE)
# 预测nnPred "class")
# 评估confusionMatrix(factor(nnPred), testTransformed$Species)

# 设置训练控制ctrl "cv", number = 5, classProbs = TRUE)
# 比较多个模型modelList   logistic = train(Species ~ ., data = trainTransformed, method = "multinom", trControl = ctrl),  rf = train(Species ~ ., data = trainTransformed, method = "rf", trControl = ctrl),  svm = train(Species ~ ., data = trainTransformed, method = "svmRadial", trControl = ctrl),  xgb = train(Species ~ ., data = trainTransformed, method = "xgbTree", trControl = ctrl))
# 比较模型性能results summary(results)
# 可视化比较dotplot(results)

# 创建多个基学习器的预测pred1 $logistic, testTransformed, type = "prob")pred2 $rf, testTransformed, type = "prob")pred3 $svm, testTransformed, type = "prob")
# 简单平均集成ensemblePred finalPred 
# 评估集成模型confusionMatrix(factor(finalPred), testTransformed$Species)


    
# 模型揭示与可视化# 特征重要性importance $rf)plot(importance)
# ROC曲线（二分类问题示例）rocCurve $Species == "setosa"),               predictor = as.numeric(pred1[, "setosa"]))plot(rocCurve, print.auc = TRUE)
# 决策边界可视化（选择两个特征）ggplot(trainTransformed, aes(x = Sepal.Length, y = Petal.Length, color = Species)) +  geom_point() +  stat_ellipse() +  ggtitle("Feature Space with Decision Boundaries")