RT Conference Proceedings
T1 Testing the Robustness of Machine Learning Models Through Mutations
A1 Méndez Hurtado, Manuel
A1 Benito Parejo, Miguel
A1 García Merayo, María De Las Mercedes
AB The reliable performance of machine learning algorithms stands as a critical and foundational concern. Usually, scientific attention centres solely on this aspect when selecting among models. However, in real-world scenarios, datasets are vulnerable to human errors during data input. Consequently, algorithms must display consistency and resilience against such errors. We assert that, especially in real-world applications, the resilience is, akin to the performance, a critical characteristic when selecting one algorithm over another. To address this concern, we propose a novel methodology for assessing model robustness by evaluating models both before and after applying mutations to the dataset. To validate the effectiveness of this methodology, we analyse five commonly used machine learning algorithms in a case study concerning traffic flow forecasting in Madrid. In assessing the robustness of the models, we introduce two metrics derived from well-known regression measurements. The results clearly reveal that the random forest model shows the highest robustness, according to our analysis, and that different models can exhibit very different behaviours in terms of this aspect.
SN 9783031702471
SN 9783031702488
SN 1865-0929
SN 1865-0937
YR 2024
FD 2024-09-08
LK https://hdl.handle.net/20.500.14352/119490
UL https://hdl.handle.net/20.500.14352/119490
LA eng
DS Docta Complutense
RD 22 abr 2025