Movatterモバイル変換

leakR

Welcome toleakR, an R package designed to helpresearchers, data scientists, and machine learning practitionersrigorously detect and diagnose data leakage in their workflows.

Data leakage is a pervasive yet often overlooked issue thatundermines the integrity and reproducibility of predictive models byallowing unintended information to “leak” between training and testingphases. leakR provides a modular, extensible toolkit for detecting themost common and impactful forms of leakage, starting with tabular datacontamination, target leakage, and temporal misalignments, while layingthe foundation for a universal leakage detection framework acrossdiverse data domains.

Installation

From CRAN (Recommended)

install.packages("leakr")

From GitHub (DevelopmentVersion)

For the latest features and bug fixes:

# Install devtools if you don't have itinstall.packages("devtools")# Install leakR from GitHubdevtools::install_github("cherylisabella/leakR")

Quick Start

library(leakr)# Basic audit of your datasetreport<-leakr_audit(iris,target ="Species")# View summary of issues foundleakr_summarise(report)# Generate diagnostic visualizationsleakr_plot(report)# Access detailed resultsprint(report)

Main Functions

Function	Purpose
`leakr_audit()`	Main auditing function - detects leakage across your dataset
`leakr_summarise()`	Generate human-readable summaries of detected issues
`leakr_plot()`	Create diagnostic visualizations highlighting problems
`leakr_from_caret()`	Import and audit caret workflow objects
`leakr_from_tidymodels()`	Import and audit tidymodels workflow objects
`leakr_from_mlr3()`	Import and audit mlr3 workflow objects

Learn More

Get started with the comprehensive vignettes:

# Getting started guidevignette("getting-started",package ="leakr")# Advanced detection techniquesvignette("advanced-detection",package ="leakr")# Framework integration examplesvignette("framework-integration",package ="leakr")

Why leakR?

Automates leakage detection, filling a keymethodological gap
Designed for clarity, reproducibility, and transparent MLresearch
Modular architecture supports gradual expansion(time series, NLP, images)
Useful for both academic and industryworkflows

What leakR Detects

Train/test contamination - Overlapping recordsbetween training and test sets
Target leakage - Features that contain informationabout the target variable that wouldn’t be available at predictiontime
Duplicate rows/records - Exact and near-duplicateobservations that can inflate performance metrics
Temporal misalignments - Time-based data leaks intime series analysis

Key Features

Visual summaries of suspicious patterns and leakagehotspots
Detailed leakage reports suitable for audits, peerreview, or publications
Clean APIs for seamless integration into existingML workflows
Example vignettes demonstrating real leakagephenomena with code illustrations
Framework integration with caret, tidymodels, andmlr3

Development Roadmap

Phase 1: Core tabular leakage detectors ✓
Phase 2: Time series leakage detection (inprogress)
Phase 3: Domain-specific extensions (NLP, imagepipelines)
Phase 4: Pipeline integration and multi-languagesupport

Citation

If you use leakR in your research, please cite:

@Manual{leakr2025,  title = {leakR: Data Leakage Detection Tools for Machine Learning},  author = {Cheryl Isabella Lim},  year = {2025},  note = {R package version 0.1.0},  url = {https://github.com/cherylisabella/leakR},}

License

This project is licensed under the MIT License - see theLICENSE file for details.

leakR is currently under development. Feedback and contributions arewelcome from the community!

[8]ページ先頭