Train regularized logistic regression in R using caret package

LogisticReg.R

## Regularized Logistic Regression in R
## Created on 12 May 2019

## Load dataset
library(mlbench)
data("BreastCancer")

## Clean data
## head(BreastCancer)
## str(BreastCancer)
mean(complete.cases(BreastCancer))
df <- na.omit(BreastCancer)
df$Id <- NULL

## Split data
set.seed(1)
id <- sample(1:nrow(df), 0.8*nrow(df))
train_df <- df[id, ]
test_df <- df[-id, ]
nrow(train_df); nrow(test_df)

## Regularization 
library(caret)
library(glmnet)

## Train with 5-Fold CV
set.seed(1)
reguarlized_model <- train(Class ~ ., data = train_df, 
                           method = "glmnet", 
                           metric = "Accuracy",
                           tuneLength = 15, ## we'll try 15 alpha and 15 lambda values
                           trControl = trainControl(method = "cv", 
                                                    number = 5,
                                                    search = "random", ## we use random search
                                                    verboseIter = T))

## Aggregating results
## Selecting tuning parameters
## Fitting alpha = 0.77, lambda = 0.021 on full training set

## Best Tuned Model
lasso_model$bestTune

## Train Accuracy
p3 <- predict(reguarlized_model, type = "prob")
p3 <- ifelse(p3[[2]] >= 0.5, T, F)
table(p3, train_df$Class)
print(sum(diag(table(p3, train_df$Class)))/ nrow(train_df))

## Test Accuracy
p4 <- predict(reguarlized_model, newdata = test_df, type = "prob")
p4 <- ifelse(p4[[2]] >= 0.5, T, F)
table(p4, test_df$Class)
print(sum(diag(table(p4, test_df$Class)))/ nrow(test_df))