TY - GEN
T1 - Abstract Interpretation-Based Feature Importance for Support Vector Machines
AU - Pal, Abhinandan
AU - Ranzato, Francesco
AU - Urban, Caterina
AU - Zanella , Marco
PY - 2024
Y1 - 2024
N2 - We study how a symbolic representation for support vector machines (SVMs) specified by means of abstract interpretation can be exploited for: (1) enhancing the interpretability of SVMs through a novel feature importance measure, called abstract feature importance (AFI), that does not depend in any way on a given dataset or the accuracy of the SVM and is very fast to compute; and (2) certifying individual fairness of SVMs and producing concrete counterexamples when this verification fails. We implemented our methodology and we empirically showed its effectiveness on SVMs based on linear and nonlinear (polynomial and radial basis function) kernels. Our experimental results prove that, independently of the accuracy of the SVM, our AFI measure correlates much strongly with stability of the SVM to feature perturbations than major feature importance measures available in machine learning software such as permutation feature importance, therefore providing better insight into the trustworthiness of SVMs.
AB - We study how a symbolic representation for support vector machines (SVMs) specified by means of abstract interpretation can be exploited for: (1) enhancing the interpretability of SVMs through a novel feature importance measure, called abstract feature importance (AFI), that does not depend in any way on a given dataset or the accuracy of the SVM and is very fast to compute; and (2) certifying individual fairness of SVMs and producing concrete counterexamples when this verification fails. We implemented our methodology and we empirically showed its effectiveness on SVMs based on linear and nonlinear (polynomial and radial basis function) kernels. Our experimental results prove that, independently of the accuracy of the SVM, our AFI measure correlates much strongly with stability of the SVM to feature perturbations than major feature importance measures available in machine learning software such as permutation feature importance, therefore providing better insight into the trustworthiness of SVMs.
UR - https://www.scopus.com/pages/publications/85182006418
U2 - 10.1007/978-3-031-50524-9_2
DO - 10.1007/978-3-031-50524-9_2
M3 - Conference contribution
AN - SCOPUS:85182006418
SN - 9783031505232
T3 - Lecture Notes in Computer Science
SP - 27
EP - 49
BT - Verification, Model Checking, and Abstract Interpretation
A2 - Dimitrova, Rayna
A2 - Lahav, Ori
A2 - Wolff, Sebastian
PB - Springer
T2 - 25th International Conference on Verification, Model Checking, and Abstract Interpretation
Y2 - 15 January 2024 through 16 January 2024
ER -