R-Classification: Predictive Power with tidymodels

Machine learning classification using the tidymodels framework

Overview

Beginner Friendly Machine Learning Classification

Dive into classification with R and the elegant tidymodels framework! Learn to build, evaluate, and refine machine learning models that predict categorical outcomes through practical exercises including a wine classification challenge.

What You’ll Learn

• 🧹 Data preprocessing with {recipes}
• 📊 Train/test splitting with {rsample}
• 🤖 Model fitting with {parsnip}
• 📈 Performance evaluation with {yardstick}
• 🎯 Complete ML workflow from data to predictions

Prerequisites

Required Knowledge:

• Basic R programming
• Understanding of classification concepts
• Familiarity with dplyr helpful

No Prior Experience Needed:

• Machine learning
• tidymodels framework

Key Packages

{tidymodels}

{recipes}

{parsnip}

{rsample}

{yardstick}

{tune}

The tidymodels Ecosystem

Core Packages

{rsample} - Data splitting and resampling

library(tidymodels)

# Split data
data_split <- initial_split(data, prop = 0.75, strata = outcome)
train_data <- training(data_split)
test_data <- testing(data_split)

{recipes} - Feature engineering

recipe_spec <- recipe(outcome ~ ., data = train_data) %>%
  step_normalize(all_numeric_predictors()) %>%
  step_dummy(all_nominal_predictors())

{parsnip} - Model specification

model_spec <- logistic_reg() %>%
  set_engine("glm") %>%
  set_mode("classification")

{workflows} - Combine recipe + model

workflow_fit <- workflow() %>%
  add_recipe(recipe_spec) %>%
  add_model(model_spec) %>%
  fit(data = train_data)

{yardstick} - Performance metrics

predictions %>%
  metrics(truth = outcome, estimate = .pred_class)

Workshop Content

1. Data Preparation

• Loading and exploring data
• Handling missing values
• Feature selection
• Train/test splitting with stratification

2. Preprocessing with Recipes

Common Steps:

• step_normalize() - Scale numeric features
• step_dummy() - Create dummy variables
• step_impute_*() - Handle missing data
• step_pca() - Dimensionality reduction
• step_interact() - Feature interactions

3. Model Building

Classification Algorithms:

• Logistic regression
• Decision trees
• Random forests
• Support vector machines
• Neural networks

Consistent Interface:

# Same syntax for different models!
log_spec <- logistic_reg() %>% set_engine("glm")
rf_spec <- rand_forest() %>% set_engine("ranger")
svm_spec <- svm_rbf() %>% set_engine("kernlab")

4. Model Evaluation

Classification Metrics:

• Accuracy
• Precision/Recall
• F1 Score
• ROC AUC
• Confusion matrix

5. Hyperparameter Tuning

# Define tuning grid
rf_spec <- rand_forest(
  mtry = tune(),
  trees = tune(),
  min_n = tune()
) %>%
  set_engine("ranger") %>%
  set_mode("classification")

# Cross-validation
cv_folds <- vfold_cv(train_data, v = 5)

# Tune
tune_results <- tune_grid(
  workflow() %>% add_recipe(recipe_spec) %>% add_model(rf_spec),
  resamples = cv_folds,
  grid = 25
)

# Select best
best_params <- select_best(tune_results, metric = "roc_auc")

Hands-On: Wine Classification

Challenge: Predict wine quality (Good/Bad) based on chemical properties.

Steps:

1. Explore wine dataset
2. Create preprocessing recipe
3. Try multiple classification algorithms
4. Evaluate and compare models
5. Select best performer
6. Make final predictions

Example Workflow:

# Load data
data(wine_quality)

# Split
wine_split <- initial_split(wine_quality, strata = quality)
wine_train <- training(wine_split)
wine_test <- testing(wine_split)

# Recipe
wine_recipe <- recipe(quality ~ ., data = wine_train) %>%
  step_normalize(all_numeric_predictors()) %>%
  step_corr(all_numeric_predictors(), threshold = 0.9)

# Model
rf_model <- rand_forest(trees = 1000) %>%
  set_engine("ranger", importance = "impurity") %>%
  set_mode("classification")

# Workflow
wine_wf <- workflow() %>%
  add_recipe(wine_recipe) %>%
  add_model(rf_model)

# Fit
wine_fit <- wine_wf %>% fit(data = wine_train)

# Predict
wine_pred <- wine_fit %>%
  predict(wine_test) %>%
  bind_cols(wine_test)

# Evaluate
wine_pred %>%
  metrics(truth = quality, estimate = .pred_class)

Practical Applications in Pharma

Clinical Trials

• Patient risk stratification
• Responder identification
• Adverse event prediction
• Site selection

Drug Discovery

• Compound classification
• Activity prediction
• Toxicity screening

Operations

• Trial enrollment prediction
• Resource allocation
• Quality control

Best Practices

✅ Do’s

• Always use train/test split
• Stratify on outcome variable
• Use cross-validation for tuning
• Check for data leakage
• Document preprocessing steps

❌ Don’ts

• Don’t peek at test set during development
• Don’t overfit with too many features
• Don’t ignore class imbalance
• Don’t forget to normalize/scale
• Don’t skip exploratory analysis

Learning Outcomes

✅ Preprocess data with {recipes}
✅ Split data properly with {rsample}
✅ Fit classification models with {parsnip}
✅ Evaluate performance with {yardstick}
✅ Tune hyperparameters systematically
✅ Build complete ML pipelines

Beyond the Workshop

Next Steps:

• Explore other algorithms (XGBoost, neural nets)
• Learn advanced feature engineering
• Study model interpretation (SHAP, LIME)
• Practice with real datasets

Resources:

• tidymodels.org
• Tidy Modeling with R (book)
• tidymodels tag on Stack Overflow

---

Similar Workshops

• Bayesian Survival Models - Applied Stan usage
• rpact Trial Design - Frequentist alternative

Related Presentations

• BayesERtools - Production Bayesian package

Next Steps

• Apply skills: Bayesian Survival workshop
• Career path: Bayesian Methods expertise

---

Last updated: November 2025 | R/Pharma 2025 Conference