CARRoT: R-package for predictive modelling by means of regression, adjusted for multiple regularisation methods.

Area of Science:

• Statistical modeling
• Machine learning
• Bioinformatics

Background:

• Predictive modeling is crucial for data analysis.
• Existing methods for model selection have limitations.
• Robust internal validation is essential for reliable predictive models.

Purpose of the Study:

• Introduce CARRoT, an R-package for predictive modeling.
• Provide a tool for exploratory data analysis and model selection.
• Compare CARRoT's performance against other model selection techniques.

Main Methods:

• CARRoT performs exhaustive search for regression models.
• It employs multiple hold-outs for internal model validation.
• Considers collinearity, event per variable rules (EPVs), R-squared, and interactions.

Main Results:

• CARRoT ranks candidate models by predictive power using cross-validation.
• It quantifies predictive performance using error, accuracy, and AUROC.
• The package facilitates parallel processing to reduce computation time.

Conclusions:

• CARRoT offers a comprehensive approach to predictive model selection.
• It provides a robust framework for evaluating and comparing different modeling strategies.
• The package aids in identifying models with superior predictive accuracy.