On the Properness of Incorporating Binary Classification Machine Learning Algorithms into Safety-Critical Systems

Research Area: Uncategorized Year: 2022
Type of Publication: Article Keywords: Machine Learning, Performance Metrics, Safety Measures, Safety-Critical Systems
Authors: Mohamad Gharib; Tommaso Zoppi; Andrea Bondavalli
Journal: IEEE Transactions on Emerging Topics in Computing
Many manufacturers are willing to incorporate Machine Learning (ML) algorithms into their systems, especially those to be considered as Safety-Critical Systems (SCS). ML algorithms that perform binary classification (i.e., binary classifiers) find a wide applicability as error detectors, intrusion detectors or failure predictors, provided that their performance complies with the safety requirements of the SCS. However, the performance analysis of binary classifiers usually relies on metrics that were not developed with safety concerns in mind and consequently may not provide meaningful evidence to decide whether it is appropriate to incorporate an binary classifier into a SCS. In this paper, we empirically assess the properness of such incorporation by analyzing the distribution of misclassifications of binary classifiers. We show that analyzing the distribution of misclassifications, instead of simply counting them, allows us to better assess the adequacy of a given binary classifier. This allows identifying areas of the classification space where the binary classifier is likely to misclassify and therefore constitutes actionable information to deal with the special nature of SCS. Our assessment takes a deeper view of the classification performance concerning safety by using new metrics that consider the proportions of predictions that are/are not considered sufficiently safe to be used by incorporating SCS. The results of our experiment allow discussing the potential of such distribution analysis for deciding if and how to properly incorporate a binary classifier into a SCS

Resilient Computing Lab, 2011

Joomla - Realizzazione siti web