ClusterValidateR: Clustered Validation and Calibration of Ranking-Based Predictive Models

Résumé fourni par la source

ClusterValidateR v0.1.0 — First Developer Preview Overview ClusterValidateR is an R package for the validation, calibration, and comparative assessment of predictive models using cluster-aware cross-validation. Many predictive modelling studies involve naturally grouped observations, such as patients within hospitals, students within schools, teams within tournaments, or locations within regions. Traditional random cross-validation often produces overly optimistic estimates of predictive performance because observations from the same cluster can appear in both the training and testing sets. ClusterValidateR addresses this problem by providing validation methods that respect the underlying clustering structure, resulting in more reliable estimates of model performance. ✨ Key Features Cluster-aware data validation Leave-one-cluster-out cross-validation Support for linear, logistic, and binomial regression models Comparative evaluation of multiple predictive models Prediction performance metrics including: Root Mean Squared Error (RMSE) Brier Score Log Loss Area Under the ROC Curve (AUC) Calibration assessment using: Calibration curves Calibration intercepts Calibration slopes Expected Calibration Error (ECE) Publication-quality visualizations Simulated clustered datasets for reproducible examples and testing Reproducible workflows for statistical research 📊 Intended Applications ClusterValidateR is designed for predictive modelling problems involving clustered or grouped data, including: ⚽ Sports Analytics 🏥 Clinical Prediction Models 🦠 Epidemiology and Public Health 🎓 Educational Assessment 🌍 Environmental and Ecological Studies 💰 Economics and Finance 📈 General Statistical and Predictive Modelling 🚀 Current Status This is the first developer preview (v0.1.0). The package provides the core framework for cluster-aware predictive model validation and is suitable for evaluation, testing, and methodological development. While the package is functional, the API and feature set may evolve in future releases. 🛠 Current Capabilities The current release includes support for: Data validation for clustered datasets Leave-one-cluster-out cross-validation Linear regression validation Logistic regression validation Binomial regression validation Model comparison Calibration assessment Discrimination assessment Cross-validation performance summaries Publication-ready graphics Simulation of clustered datasets 📖 Citation If you use ClusterValidateR in your research, please cite the package using the citation information provided within R: citation("ClusterValidateR") 📄 License This project is released under the MIT License. 👥 Authors Taiwo M. Adegoke Department of Statistics ORCID: https://orcid.org/0000-0003-2507-4973 Oladapo M. Oladoja ORCID: https://orcid.org/0000-0002-0766-2248 💡 Vision ClusterValidateR aims to become a comprehensive framework for the validation, calibration, and comparative assessment of predictive models with clustered data. By promoting statistically rigorous validation strategies that respect natural grouping structures, the package seeks to improve the reproducibility, transparency, and reliability of predictive modelling across a broad range of scientific disciplines.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.