bibtype C - Conference Paper (international conference)
ARLID 0642816
utime 20260123171641.2
mtime 20251209235959.9
SCOPUS 105023988800
DOI 10.1109/IJCNN64981.2025.11227341
title (primary) (eng) Local Sensitivity Analysis of Highly Robust Regression Estimators
specification
page_count 8 s.
media_type P
serial
ARLID cav_un_epca*0642775
ISBN 979-8-3315-1042-8
title IJCNN 2025: International Joint Conference on Neural Networks Conference Proceedings
publisher
place Piscataway
name IEEE
year 2025
keyword outliers
keyword robustness
keyword linear regression
keyword implicit weighting
keyword local sensitivity
keyword subsample sensitivity
author (primary)
ARLID cav_un_auth*0345793
name1 Kalina
name2 Jan
institution UTIA-B
full_dept (cz) Stochastická informatika
full_dept (eng) Department of Stochastic Informatics
department (cz) SI
department (eng) SI
full_dept Department of Stochastic Informatics
fullinstit Ústav teorie informace a automatizace AV ČR, v. v. i.
source
url http://library.utia.cas.cz/separaty/2025/SI/kalina-0642816.pdf
cas_special
abstract (eng) Robust regression methods are essential for estimating parameters in linear regression models, especially when data is contaminated by outliers or small fluctuations, common in real-world applications. This paper investigates the local sensitivity of robust regression estimators, focusing on how small modifications in the data affect their predictions. Local sensitivity analysis (LSA) offers valuable insights into the robustness of regression models, a crucial property in decision-making fields like economics, finance, and engineering, where data integrity is often compromised. We examine robust estimators based on implicit weights, including the least trimmed squares (LTS) and least weighted squares (LWS) estimators, along with their regularized versions. A novel adaptive version of the regularized LWS estimator is proposed, incorporating data-driven weights. Experiments on publicly available datasets show that while the LTS and LTS-lasso estimators exhibit high local sensitivity, LWS and LWS-lasso estimators—especially with adaptive weights—demonstrate superior robustness and predictive performance. These findings highlight the importance of considering local sensitivity in robust regression models, particularly in economic data, where small fluctuations can significantly impact predictions and decisions. The importance of local sensitivity for machine learning is also discussed, with suggestions for future applications in robust machine learning models, particularly in economic contexts.
action
ARLID cav_un_auth*0498693
name International Joint Conference on Neural Networks 2025 (IJCNN 2025)
dates 20250630
mrcbC20-s 20250705
place Rome
country IT
RIV IN
FORD0 10000
FORD1 10200
FORD2 10201
reportyear 2026
presentation_type PR
inst_support RVO:67985556
permalink https://hdl.handle.net/11104/0372672
mrcbC61 1
confidential S
arlyear 2025
mrcbU14 105023988800 SCOPUS
mrcbU24 PUBMED
mrcbU34 WOS
mrcbU63 cav_un_epca*0642775 IJCNN 2025: International Joint Conference on Neural Networks Conference Proceedings IEEE 2025 Piscataway 979-8-3315-1042-8