binaryClass: R Package for Binary Classification

Overview

The binaryClass package provides a comprehensive toolkit for binary classification tasks in R. It implements powerful functions for model training, evaluation, and prediction, with a focus on comparing different classification approaches.

Installation

From GitHub Release

# Install the released version from GitHub
install.packages("https://codoom1.github.io/binaryClass/binaryClass_1.0.0.tar.gz", repos = NULL)

From GitHub with devtools

# Install development version
remotes::install_github("codoom1/binaryClass")

Key Features

• OptimalModelSearch: Automatically compares multiple binary classification models and selects the best model based on AUC, Accuracy, or AIC. Models compared include:

  • Full Logistic Regression
  • Backward Stepwise Selection
  • Forward Stepwise Selection
  • GAM (Generalized Additive Model)
  • Lasso Regression
  • Ridge Regression

• Visualization Options:

  • ROC Curves: For AUC criterion - visualize model discrimination ability
    • Automatic plotting with OptimalModelSearch(..., criterion="AUC", plot_roc=TRUE)
    • Comparison plotting with plot_model_rocs()
  • Confusion Matrices: For Accuracy criterion - visualize classification performance
    • Automatic plotting with OptimalModelSearch(..., criterion="Accuracy", plot_cm=TRUE)
    • Manual plotting with plot_model_cm()

• Additional Tools:

  • compare_model_rocs: Standalone function to compare ROC curves for stepwise, lasso, and ridge models
  • plot_descrip: Create descriptive visualizations for binary classification datasets

Usage Examples

Model Selection

# Load example data
library(binaryClass)
library(mlbench)
data(PimaIndiansDiabetes)
df <- PimaIndiansDiabetes
df$diabetes <- ifelse(df$diabetes=="neg", 0, 1)

# Find best model using AUC criterion with ROC curve
result_auc <- OptimalModelSearch(formula=diabetes~., data=df, 
                                criterion="AUC", plot_roc=TRUE,
                                suppress_warnings=TRUE)

# Find best model using Accuracy criterion with confusion matrix
result_acc <- OptimalModelSearch(formula=diabetes~., data=df, 
                                criterion="Accuracy", plot_cm=TRUE)

# Find best model using AIC criterion
result_aic <- OptimalModelSearch(formula=diabetes~., data=df, 
                                criterion="AIC")

Extract and Use the Best Model

# Get the best model from results
best_model <- extract_best_model(result_auc)

# Use it for predictions on new data
new_pred <- predict(best_model, newdata=new_data)

Visualizations

# ROC curves for AUC criterion
plot_model_rocs(result_auc, comparison=TRUE)

# Confusion matrix for Accuracy criterion
plot_model_cm(result_acc)

# Standalone ROC curve comparison
compare_model_rocs(formula=diabetes~., data=df)

Dependencies

• stats
• MASS
• gam
• glmnet
• pROC
• caret
• e1071
• graphics
• grDevices

License

This package is licensed under the MIT License.

Contributing

Contributions to improve binaryClass are welcome. Please feel free to submit a pull request or open an issue on GitHub.