Related Experiment Video
Updated: Jul 5, 2025
08:51
Statistical Modelling of Cortical Connectivity Using Non-invasive Electroencephalograms
Published on: November 1, 2019
5.6K
The effects of data leakage on connectome-based machine learning models.
Matthew Rosenblatt1, Link Tejavibulya2, Rongtao Jiang3
1Department of Biomedical Engineering, Yale University, New Haven, CT.
Biorxiv : the Preprint Server for Biology
|January 18, 2024
Summary
Data leakage in neuroimaging predictive models can inflate performance, especially with feature selection or repeated subjects. Avoiding leakage is crucial for valid and reproducible brain-behavior relationship studies.
Area of Science:
- Neuroimaging
- Machine Learning
- Brain-Behavior Relationships
Background:
- Predictive modeling is key in neuroimaging for understanding brain-behavior links.
- Data leakage, or unintentional information breach between training and testing sets, compromises model validity.
- Previous research indicates leakage is common in machine learning, but its specific impact on neuroimaging is less understood.
Conclusions:
- Data leakage has variable effects on neuroimaging predictive models, ranging from performance inflation to decrease.
- Understanding these effects highlights the critical need for rigorous data handling to ensure the validity and reproducibility of neuroimaging research.
- Avoiding data leakage is paramount for accurate identification of brain-behavior relationships and reliable generalizability of findings.
