An Application of Quadrature-Inspired Generalized Choquet Integral to Anomaly Detection and Multi-Class Problems

作     者：Karczmarek, Pawel Dolecki, Michal Powroznik, Pawel Lagodowski, Zbigniew A. Gregosiewicz, Adam Galka, Lukasz Pedrycz, Witold Czerwinski, Dariusz 

作者机构：Department of Computer Science Lublin University of Technology ul. Nadbystrzycka 36B Lublin20-618 Poland Department of Mathematics Lublin University of Technology ul. Nadbystrzycka 36B Lublin20-618 Poland Department of Electrical & Computer Engineering University of Alberta EdmontonT6R 2V4 AB Canada Department of Electrical and Computer Engineering King Abdulaziz University Jeddah21589 Saudi Arabia Systems Research Institute Polish Academy of Sciences Warsaw Poland 

出 版 物：《SSRN》 

年 卷 期：2023年

核心收录：

主　　题：Anomaly detection 

摘      要：In this study, we thoroughly analyze and evaluate a novel approach to derive an aggregate classification score. The aggregation process applied to various classifiers is realized on a basis of Choquet integral enhancements. These new expressions are inspired by Newton-Cotes quadratures and other formulae well-known from numerical analysis. Unlike previous approaches where to find the Choquet integral generalization one has to use two or three adjacent values associated to the membership of a given element to individual classes, here, one can use more efficient enhancement. Namely, a t-norm appearing under the symbol of integral can be effectively replaced by mathematical expressions used in carrying out numerical integration formulae. It yields more precise results and fully reflects the concept of numerical integration. Moreover, in a series of experiments, we thoroughly analyze the performance of the proposed approach in terms of accuracy in the classification tasks. We analyze the pros and cons of the new approach and establish the experimental settings which can be used in similar tasks. © 2023, The Authors. All rights reserved.