【R】如何确定最适合数据集的机器学习算法

抽查(Spot checking)机器学习算法是指如何找出最适合于给定数据集的算法模型。本文中我将介绍八个常用于抽查的机器学习算法，文中还包括各个算法的 R 语言代码，你可以将其保存并运用到下一个机器学习项目中。

适用于你的数据集的最佳算法

抽查哪些算法？

首先，你可以思考哪些算法可能适用于你的数据集。其次，我建议尽可能地尝试混合算法并观察哪个方法最适用于你的数据集。

让我们具体看下如何实现这几个想法。下一章中我们将看到如何在 R 语言中实现相应的机器学习算法。

如何在 R 语言中抽查算法？

1.常规的训练和预测方法2.caret包的用法

你需要知道给定算法对应的软件包和函数，同时你还需了解如何利用caret包实现这些常用的算法，从而你可以利用caret包的预处理、算法评估和参数调优的能力高效地评估算法的精度。本文中将用到两个标准的数据集：

1.回归模型：BHD(Boston Housing Dataset)2.分类模型: PIDD(Pima Indians Diabetes Dataset)

本文中的算法将被分成两组进行介绍：

1.线性算法：简单、较大的偏倚、运算速度快2.非线性算法：复杂、较大的方差、高精确度

下文中的所有代码都是完整的，因此你可以将其保存下来并运用到下个机器学习项目中。

线性算法

这类方法对模型的函数形式有严格的假设条件，虽然这些方法的运算速度快，但是其结果偏倚较大。这类模型的最终结果通常易于解读，因此如果线性模型的结果足够精确，那么你没有必要采用较为复杂的非线性模型。

线性回归模型

stat包中的lm()函数可以利用最小二乘估计拟合线性回归模型。

# load the library  library(mlbench)  # load data  data(BostonHousing)  # fit model  fit <- lm(mdev~>, BostonHousing)  # summarize the fit  print(fit)  # make predictions  predictions <- predict(fit, BostonHousing)  # summarize accuracy  mse <- mean((BostonHousing$medv - predictions)^2)  print(mse)    # caret  # load libraries  library(caret)  library(mlbench)  # load dataset  data(BostonHousing)  # train  set.seed(7)  control <- trainControl(method="cv", number=5)  fit.lm <- train(medv~., data=BostonHousing, method="lm", metric="RMSE", preProc=c("center", "scale"), trControl=control)  # summarize fit  print(fit.lm)

罗吉斯回归模型

stat包中glm()函数可以用于拟合广义线性模型。它可以用于拟合处理二元分类问题的罗吉斯回归模型。

# load the library  library(mlbench)  # Load the dataset  data(PimaIndiansDiabetes)  # fit model  fit <- glm(diabetes~., data=PimaIndiansDiabetes, family=binomial(link='logit'))  # summarize the fit  print(fit)  # make predictions  probabilities <- predict(fit, PimaIndiansDiabetes[,1:8], type='response')  predictions <- ifelse(probabilities > 0.5,'pos','neg')  # summarize accuracy  table(predictions, PimaIndiansDiabetes$diabetes)    # caret  # load libraries  library(caret)  library(mlbench)  # Load the dataset  data(PimaIndiansDiabetes)  # train  set.seed(7)  control <- trainControl(method="cv", number=5)  fit.glm <- train(diabetes~., data=PimaIndiansDiabetes, method="glm", metric="Accuracy", preProc=c("center", "scale"), trControl=control)  # summarize fit  print(fit.glm)

线性判别分析

MASS包中的lda()函数可以用于拟合线性判别分析模型。

# load the libraries  library(MASS)  library(mlbench)  # Load the dataset  data(PimaIndiansDiabetes)  # fit model  fit <- lda(diabetes~., data=PimaIndiansDiabetes)  # summarize the fit  print(fit)  # make predictions  predictions <- predict(fit, PimaIndiansDiabetes[,1:8])$class  # summarize accuracy  table(predictions, PimaIndiansDiabetes$diabetes)    # caret  # load libraries  library(caret)  library(mlbench)  # Load the dataset  data(PimaIndiansDiabetes)  # train  set.seed(7)  control <- trainControl(method="cv", number=5)  fit.lda <- train(diabetes~., data=PimaIndiansDiabetes, method="lda", metric="Accuracy", preProc=c("center", "scale"), trControl=control)  # summarize fit  print(fit.lda)

正则化回归

glmnet包中的glmnet()函数可以用于拟合正则化分类或回归模型。分类模型：

# load the library  library(glmnet)  library(mlbench)  # load data  data(PimaIndiansDiabetes)  x <- as.matrix(PimaIndiansDiabetes[,1:8])  y <- as.matrix(PimaIndiansDiabetes[,9])  # fit model  fit <- glmnet(x, y, family="binomial", alpha=0.5, lambda=0.001)  # summarize the fit  print(fit)  # make predictions  predictions <- predict(fit, x, type="class")  # summarize accuracy  table(predictions, PimaIndiansDiabetes$diabetes)    # caret  # load libraries  library(caret)  library(mlbench)  library(glmnet)  # Load the dataset  data(PimaIndiansDiabetes)  # train  set.seed(7)  control <- trainControl(method="cv", number=5)  fit.glmnet <- train(diabetes~., data=PimaIndiansDiabetes, method="glmnet", metric="Accuracy", preProc=c("center", "scale"), trControl=control)  # summarize fit  print(fit.glmnet)

回归模型：

# load the libraries  library(glmnet)  library(mlbench)  # load data  data(BostonHousing)  BostonHousing$chas <- as.numeric(as.character(BostonHousing$chas))  x <- as.matrix(BostonHousing[,1:13])  y <- as.matrix(BostonHousing[,14])  # fit model  fit <- glmnet(x, y, family="gaussian", alpha=0.5, lambda=0.001)  # summarize the fit  print(fit)  # make predictions  predictions <- predict(fit, x, type="link")  # summarize accuracy  mse <- mean((y - predictions)^2)  print(mse)    # caret  # load libraries  library(caret)  library(mlbench)  library(glmnet)  # Load the dataset  data(BostonHousing)  # train  set.seed(7)  control <- trainControl(method="cv", number=5)  fit.glmnet <- train(medv~., data=BostonHousing, method="glmnet", metric="RMSE", preProc=c("center", "scale"), trControl=control)  # summarize fit  print(fit.glmnet)

非线性算法

# knn direct classification    # load the libraries  library(caret)  library(mlbench)  # Load the dataset  data(PimaIndiansDiabetes)  # fit model  fit <- knn3(diabetes~., data=PimaIndiansDiabetes, k=3)  # summarize the fit  print(fit)  # make predictions  predictions <- predict(fit, PimaIndiansDiabetes[,1:8], type="class")  # summarize accuracy  table(predictions, PimaIndiansDiabetes$diabetes)    # caret  # load libraries  library(caret)  library(mlbench)  # Load the dataset  data(PimaIndiansDiabetes)  # train  set.seed(7)  control <- trainControl(method="cv", number=5)  fit.knn <- train(diabetes~., data=PimaIndiansDiabetes, method="knn", metric="Accuracy", preProc=c("center", "scale"), trControl=control)  # summarize fit  print(fit.knn)

回归模型：

# load the libraries  library(caret)  library(mlbench)  # load data  data(BostonHousing)  BostonHousing$chas <- as.numeric(as.character(BostonHousing$chas))  x <- as.matrix(BostonHousing[,1:13])  y <- as.matrix(BostonHousing[,14])  # fit model  fit <- knnreg(x, y, k=3)  # summarize the fit  print(fit)  # make predictions  predictions <- predict(fit, x)  # summarize accuracy  mse <- mean((BostonHousing$medv - predictions)^2)  print(mse)    # caret  # load libraries  library(caret)  data(BostonHousing)  # Load the dataset  data(BostonHousing)  # train  set.seed(7)  control <- trainControl(method="cv", number=5)  fit.knn <- train(medv~., data=BostonHousing, method="knn", metric="RMSE", preProc=c("center", "scale"), trControl=control)  # summarize fit  print(fit.knn)

朴素贝叶斯算法e1071包中的naiveBayes()函数可用于拟合分类问题中的朴素贝叶斯模型。

# load the libraries  library(e1071)  library(mlbench)  # Load the dataset  data(PimaIndiansDiabetes)  # fit model  fit <- naiveBayes(diabetes~., data=PimaIndiansDiabetes)  # summarize the fit  print(fit)  # make predictions  predictions <- predict(fit, PimaIndiansDiabetes[,1:8])  # summarize accuracy  table(predictions, PimaIndiansDiabetes$diabetes)    # caret  # load libraries  library(caret)  library(mlbench)  # Load the dataset  data(PimaIndiansDiabetes)  # train  set.seed(7)  control <- trainControl(method="cv", number=5)  fit.nb <- train(diabetes~., data=PimaIndiansDiabetes, method="nb", metric="Accuracy", trControl=control)  # summarize fit  print(fit.nb)

# Classification Example:  # load the libraries  library(kernlab)  library(mlbench)  # Load the dataset  data(PimaIndiansDiabetes)  # fit model  fit <- ksvm(diabetes~., data=PimaIndiansDiabetes, kernel="rbfdot")  # summarize the fit  print(fit)  # make predictions  predictions <- predict(fit, PimaIndiansDiabetes[,1:8], type="response")  # summarize accuracy  table(predictions, PimaIndiansDiabetes$diabetes)    # caret  # load libraries  library(caret)  library(mlbench)  # Load the dataset  data(PimaIndiansDiabetes)  # train  set.seed(7)  control <- trainControl(method="cv", number=5)  fit.svmRadial <- train(diabetes~., data=PimaIndiansDiabetes, method="svmRadial", metric="Accuracy", trControl=control)  # summarize fit  print(fit.svmRadial)

回归模型：

# Regression Example:  # load the libraries  library(kernlab)  library(mlbench)  # load data  data(BostonHousing)  # fit model  fit <- ksvm(medv~., BostonHousing, kernel="rbfdot")  # summarize the fit  print(fit)  # make predictions  predictions <- predict(fit, BostonHousing)  # summarize accuracy  mse <- mean((BostonHousing$medv - predictions)^2)  print(mse)    # caret  # load libraries  library(caret)  library(mlbench)  # Load the dataset  data(BostonHousing)  # train  set.seed(7)  control <- trainControl(method="cv", number=5)  fit.svmRadial <- train(medv~., data=BostonHousing, method="svmRadial", metric="RMSE", trControl=control)  # summarize fit  print(fit.svmRadial)

分类和回归树

rpart包中的rpart()函数可用于拟合CART分类树和回归树模型。分类模型：

# load the libraries  library(rpart)  library(mlbench)  # Load the dataset  data(PimaIndiansDiabetes)  # fit model  fit <- rpart(diabetes~., data=PimaIndiansDiabetes)  # summarize the fit  print(fit)  # make predictions  predictions <- predict(fit, PimaIndiansDiabetes[,1:8], type="class")  # summarize accuracy  table(predictions, PimaIndiansDiabetes$diabetes)    # caret  # load libraries  library(caret)  library(mlbench)  # Load the dataset  data(PimaIndiansDiabetes)  # train  set.seed(7)  control <- trainControl(method="cv", number=5)  fit.rpart <- train(diabetes~., data=PimaIndiansDiabetes, method="rpart", metric="Accuracy", trControl=control)  # summarize fit  print(fit.rpart)

回归模型：

# load the libraries  library(rpart)  library(mlbench)  # load data  data(BostonHousing)  # fit model  fit <- rpart(medv~., data=BostonHousing, control=rpart.control(minsplit=5))  # summarize the fit  print(fit)  # make predictions  predictions <- predict(fit, BostonHousing[,1:13])  # summarize accuracy  mse <- mean((BostonHousing$medv - predictions)^2)  print(mse)    # caret  # load libraries  library(caret)  library(mlbench)  # Load the dataset  data(BostonHousing)  # train  set.seed(7)  control <- trainControl(method="cv", number=2)  fit.rpart <- train(medv~., data=BostonHousing, method="rpart", metric="RMSE", trControl=control)  # summarize fit  print(fit.rpart)

其他算法

总结

本文中介绍了八个常用的机器学习算法：

1. 线性回归模型
2. 罗吉斯回归模型
3. 线性判别分析
4. 正则化回归
5. k近邻
6. 朴素贝叶斯
7. 支持向量机
8. 分类和回归树

从上文的介绍中，你可以学到如何利用 R 语言中的包和函数实现这些算法。同时你还可以学会如何利用caret包实现上文提到的所有机器学习算法。最后，你还可以将这些算法运用到你的机器学习项目中。

你的下一步计划？

你有没有试验过本文中的算法代码？

</div>