4.Using Feature Engineering Methods to Improve Model Performance.Rmd

---
output:
  word_document: default
  html_document: default
---
```{r}
#install.packages("tidymodels")
#install.packages("catdata")
#install.packages("dplyr")
library(tidymodels)
library(catdata)
library(dplyr)
data(heart)
data <- data.frame(heart)
head(data,10)
```

```{r}
data$y <- as.factor(data$y)
data$famhist <- as.factor(data$famhist)
str(data)
summary(data)
```
```{r}
set.seed(2380) # for reproducebility
# data splitting
data_split <- initial_split(data, prop = 0.80, strata = y)
# creating train set
data_train <- data_split %>%
training()
# creating test set
data_test <- data_split %>%
testing()
# dimension of train and test set
dim(data_train)
dim(data_test)
table(data_train$y);table(data_test$y)
```


```{r}
logistic_model <- logistic_reg() %>%
set_engine('glm') %>% set_mode('classification')
# Fitting a linear regression model
logistic_fit <- logistic_model %>%
fit(y ~ ., data = data_train)
# Obtaining the estimated parameters
tidy(logistic_fit)
```


FOR TRAIN


```{r}
# Estimated labels
data_labels_train <- logistic_fit %>%
predict(new_data = data_train, type = 'class')
head(data_labels_train, 5)
# Estimated probabilities
data_preds_train <- logistic_fit %>%
predict(new_data = data_train, type = 'prob')
head(data_preds_train, 5)
```


```{r}
data_results_train <- data_train %>%
  dplyr::select(y) %>%
  bind_cols(data_preds_train, data_labels_train)
head(data_results_train,10)
# Confusion matrix
conf_mat(data_results_train, truth = y, estimate = .pred_class) %>%
summary()
```



FOR TEST

```{r}
# Estimated labels
data_labels_test <- logistic_fit %>%
predict(new_data = data_test, type = 'class')
head(data_labels_test, 5)
# Estimated probabilities
data_preds_test <- logistic_fit %>%
predict(new_data = data_test, type = 'prob')
head(data_preds_test, 5)
```

```{r}
data_results_test <- data_test %>%
  dplyr::select(y) %>%
  bind_cols(data_preds_test, data_labels_test)
head(data_results_test,10)
# Confusion matrix
conf_mat(data_results_test, truth = y, estimate = .pred_class) %>%
summary()
```

```{r}
# heatmap of the confusion matrix
conf_mat(data_results_test, truth = y, estimate = .pred_class) %>%
  autoplot(type = 'heatmap')
conf_mat(data_results_train, truth = y, estimate = .pred_class) %>%
  autoplot(type = 'heatmap')
```

```{r}
# Visualizing performance across thresholds
data_results_test %>%
  roc_curve(truth = y, estimate = .pred_0) %>%
  autoplot()
data_results_train %>%
  roc_curve(truth = y, estimate = .pred_0) %>%
  autoplot()
```

```{r}
# Calculating ROC AUC
roc_auc(data_results_test, truth = y, .pred_0)
roc_auc(data_results_train, truth = y, .pred_0)
```



AUTOMATING



FEATURE ENGINEERING
FOR TEST

```{r}
data_recipe <- recipe(y ~ ., data = data_train) %>%
  # Removed correlated predictors
  step_corr(all_numeric(), threshold = 0.8) %>%
  # Normalize numeric predictors
  step_normalize(all_numeric()) %>%
  # Create dummy variables
  step_dummy(all_nominal(), -all_outcomes())
# Train recipe
data_recipe_prep <- data_recipe %>%
  prep(training = data_train)
# Transform training data
data_train_prep <-data_recipe_prep %>%
  bake(new_data = data_train)
# Transform test data
data_test_prep <- data_recipe_prep %>%
  bake(new_data = data_test)
```


```{r}
# Train logistic model
logistic_fit <- logistic_model %>%
  fit(y ~., data = data_train_prep)

# Obtain class predictions
class_preds <- predict(logistic_fit, new_data = data_test_prep,type = 'class')

# Obtain estimated probabilities
prob_preds <- predict(logistic_fit, new_data = data_test_prep,type = 'prob')

# Combine test set results
data_results_feature <- data_test_prep %>%
  dplyr::select(y) %>%
  bind_cols(class_preds, prob_preds)
# calculating accuracy
conf_mat(data_results_feature, truth = y, estimate = .pred_class) %>%
summary()
# Visualizing performance across thresholds
data_results_feature %>%
  roc_curve(truth = y, estimate = .pred_0) %>%
  autoplot()
# Calculating ROC AUC
roc_auc(data_results_feature, truth = y, .pred_0)
```

FOR TRAIN

```{r}

# Obtain class predictions
class_preds_train <- predict(logistic_fit, new_data = data_train_prep,type = 'class')

# Obtain estimated probabilities
prob_preds_train <- predict(logistic_fit, new_data = data_train_prep,type = 'prob')

# Combine test set results
data_results_feature_train <- data_train_prep %>%
  dplyr::select(y) %>%
  bind_cols(class_preds_train, prob_preds_train)
# calculating accuracy
conf_mat(data_results_feature_train, truth = y, estimate = .pred_class) %>%
summary()
# Visualizing performance across thresholds
data_results_feature_train %>%
  roc_curve(truth = y, estimate = .pred_0) %>%
  autoplot()
# Calculating ROC AUC
roc_auc(data_results_feature_train, truth = y, .pred_0)
```

```{r}
acc_before_test <- accuracy(data_results_test,truth = y, estimate = .pred_class)
acc_before_train <- accuracy(data_results_train,truth = y, estimate = .pred_class)
spc_before_test <- spec(data_results_test,truth = y, estimate = .pred_class)
spc_before_train <- spec(data_results_train,truth = y, estimate = .pred_class)
sen_before_test <- sens(data_results_test,truth = y, estimate = .pred_class)
sen_before_train <- sens(data_results_train,truth = y, estimate = .pred_class)
auc_before_test <- roc_auc(data_results_test, truth = y, .pred_0)
auc_before_train <- roc_auc(data_results_train, truth = y, .pred_0)
acc_after_test <- accuracy(data_results_feature,truth = y, estimate = .pred_class)
acc_after_train <- accuracy(data_results_feature_train,truth = y, estimate = .pred_class)
spc_after_test <- spec(data_results_feature,truth = y, estimate = .pred_class)
spc_after_train <- spec(data_results_feature_train,truth = y, estimate = .pred_class)
sen_after_test <- sens(data_results_feature,truth = y, estimate = .pred_class)
sen_after_train <- sens(data_results_feature_train,truth = y, estimate = .pred_class)
auc_after_test <- roc_auc(data_results_feature, truth = y, .pred_0)
auc_after_train <- roc_auc(data_results_feature_train, truth = y, .pred_0)
comp <- data.frame( "Before"=c(acc_before_test$.estimate,acc_before_train$.estimate,spc_before_test$.estimate,spc_before_train$.estimate,sen_before_test$.estimate,sen_before_train$.estimate,auc_before_test$.estimate,auc_before_train$.estimate),
                    "After"=c(acc_after_test$.estimate,acc_after_train$.estimate,spc_after_test$.estimate,spc_after_train$.estimate,sen_after_test$.estimate,sen_after_train$.estimate,auc_after_test$.estimate,auc_after_train$.estimate),
                    row.names = c("Test Accuracy","Train Accuracy","Test Specificity", "Train Specificity", "Test Sensitivity","Train Sensitivity", "Test Area Under Curve", "Train Area Under Curve")
)
comp
```