On Leakage in Machine Learning Pipelines

Sasse, Leonard; Raimondo, Federico; Kulkarni, Abhijit; Nicolaisen-Sobesky, Eliana; Eickhoff, Simon B.; Patil, Kaustubh R.; Love, Bradley C.; Hamdan, Sami; Lahnakoski, Juha; Dukart, Jürgen; Götz, Michael; Komeyer, Vera

doi:10.48550/ARXIV.2311.04179

Preprint

FZJ-2023-04636

On Leakage in Machine Learning Pipelines

Sasse, L.FZJ* ; Nicolaisen-Sobesky, E. ; Dukart, J.FZJ* ; Eickhoff, S. B.FZJ* ; Götz, M. ; Hamdan, S.FZJ* ; Komeyer, V.FZJ* ; Kulkarni, A. ; Lahnakoski, J.FZJ* ; Love, B. C. ; Raimondo, F.FZJ* ; Patil, K. R. (Corresponding author)FZJ*

2023
arXiv

arXiv (2023) [10.48550/ARXIV.2311.04179]

This record in other databases:

Please use a persistent id in citations: doi:10.48550/ARXIV.2311.04179 doi:10.34734/FZJ-2023-04636

Abstract: Machine learning (ML) provides powerful tools for predictive modeling. ML's popularity stems from the promise of sample-level prediction with applications across a variety of fields from physics and marketing to healthcare. However, if not properly implemented and evaluated, ML pipelines may contain leakage typically resulting in overoptimistic performance estimates and failure to generalize to new data. This can have severe negative financial and societal implications. Our aim is to expand understanding associated with causes leading to leakage when designing, implementing, and evaluating ML pipelines. Illustrated by concrete examples, we provide a comprehensive overview and discussion of various types of leakage that may arise in ML pipelines.

Keyword(s): Machine Learning (cs.LG) ; Artificial Intelligence (cs.AI) ; FOS: Computer and information sciences

Contributing Institute(s):