Skip to contents

LogisticEnsembles

The goal of LogisticEnsembles is to perform a complete analysis of logistic data. The package automatically returns 36 models (23 individual and 13 ensembles of models)

Installation

You can install the development version of LogisticEnsembles like so:

devtools::install_github("InfiniteCuriosity/LogisticEnsembles")

Example

This is a basic example which shows you how to solve a common problem:

library(LogisticEnsembles)
Logistic(data = SAHeart,
    colnum = 10,
    numresamples = 2,
    how_to_handle_strongs = 1,
    do_you_have_new_data = "N",
    save_all_trained_models = "N",
    remove_ensemble_correlations_greater_than = 1.00,
    use_parallel = "N",
    train_amount = 0.60,
    test_amount = 0.20,
    validation_amount = 0.20)

Each of the 36 models returns a probability between 0 and 1. Each of the 36 models fit the data to the training set, make predictions and measure accuracy on the test and validation sets.

The list of 36 logistic models:

  1. ADA Boost
  2. Bagged Random Forest
  3. Bagging
  4. BayesGLM
  5. BayesRNN
  6. C50
  7. Cubist
  8. Ensemble ADA Boost
  9. Ensemble Bagging
  10. Ensemble C50
  11. Ensemble Gradient Boosted
  12. Ensemble Partial Least Squares
  13. Ensemble Penalized Discrminant Analysis
  14. Ensemble Random Forest
  15. Ensemble Ranger
  16. Ensemble Regularized Discrminant Analysis
  17. Ensemble RPart
  18. Ensemble Support Vector Machines
  19. Ensemble Trees
  20. Ensemble XGBoost
  21. Flexible Discriminant Analysis
  22. Generalized Additive Models
  23. Generalized Linear Models
  24. Gradient Boosted
  25. Linear Discrmininant Analysis
  26. Linear Model
  27. Mixed Discrmininant Analysis
  28. Naive Bayes
  29. Penalized Discrminant Analysis
  30. Quadratic Discrmininant Analysis
  31. Random Forest
  32. Ranger
  33. RPart
  34. Support Vector Machines
  35. Trees
  36. XGBoost

The 13 plots automatically created by the package are:

  1. Target vs each feature (multiple barcharts)
  2. Boxplots of the numeric data
  3. Over or underfitting barchart
  4. Duration barchart
  5. Overfitting by model and resample
  6. Model accuracy barchart
  7. Accuracy by model and resample
  8. Accuracy by model
  9. ROC curves
  10. Pairwise scatterplots
  11. Correlation of the data as circles by color and size
  12. Correlation of the data by color and number

The tables and reports automtically created: 1. Summary report. This includes the Model, Accuracy, True Postive, True Negative, False Positive, False Negative, Positive Predictive Value, Negative Predictive Value, F1 score, Area under the curve, overfitting min, overfitting mean, overfitting max, and duration. 2. Data summary 3. Head of the ensemble 4. Correlation of the ensemble 5. Correlation of the data 6. Head of the data frame

The package also returns all 36 summary reports, alphabetical by model.