Fitting prediction rule ensembles to psychological research data: An introduction and tutorial.

Area of Science:

• Statistical Learning
• Machine Learning
• Psychological Research Methods

Background:

• Traditional decision tree ensembles often lack interpretability.
• There is a need for statistical learning methods that balance predictive performance with model simplicity.

Purpose of the Study:

• Introduce Prediction Rule Ensembles (PREs) as a method balancing predictive performance and interpretability.
• Demonstrate the application of the R package 'pre' for fitting PREs using psychological research data.

Main Methods:

• Utilizes decision tree ensembles (e.g., boosted trees, random forests) as a foundation.
• Extracts a subset of interpretable rules (if [condition] then [prediction]) from ensembles.
• Employs the R package 'pre' for model fitting and analysis.

Main Results:

• PREs achieve predictive performance comparable to full ensembles but with significantly reduced complexity.
• The R package 'pre' supports diverse response types (categorical, count) and advanced features.
• Demonstrated utility in psychological research contexts with real-data examples.

Conclusions:

• PREs provide a valuable approach for enhancing model interpretability in statistical learning.
• The 'pre' R package offers a practical tool for applying PREs in psychological research.
• The methodology is suitable for complex datasets and various response variable types.