Learning and Intelligent Optimization - 17th International Conference, LION 17, Nice, France, June 4-8, 2023, Revised Selected Papers
Learning and Intelligent Optimization - 17th International Conference, LION 17, Nice, France, June 4-8, 2023, Revised Selected Papers
复制标题
学习与智能优化 - 第 17 届国际会议，LION 17，法国尼斯，2023 年 6 月 4-8 日，修订后的精选论文
DOI：
10.1007/978-3-031-44505-7_2
复制
发表时间：
2023
期刊：
影响因子：
--
通讯作者：
Liapis G
中科院分区：
文献类型：
--
作者：
Liapis G
分类是机器学习研究领域的重点研究课题。可解释的机器学习算法一直在与黑盒模型抗衡，因为人们想要了解决策过程。基于数学规划的分类器受到了人们的关注，因为它们在准确性和可解释性方面可以与最先进的算法竞争。本文介绍了一种单层超盒分类方法，该方法在数学上表示为混合整数线性规划模型。它的目标是使用超盒表示来识别数据集的模式。超级盒子包含尽可能多的相应类的样本。同时，它们不允许与不同等级的超级盒子重叠。该方法的可解释性源于这样一个事实，即可以轻松地生成IF-THEN规则。为了评估所提出的方法的性能，在许多真实数据集中将其预测精度与其他最先进的可解释方法进行了比较。实验结果表明，该算法在性能上优于已有的同类算法。
Classification constitutes focal topic of study within the machine learning research community. Interpretable machine learning algorithms have been gaining ground against black box models because people want to understand the decision-making process. Mathematical programming based classifiers have received attention because they can compete with state-of-the-art algorithms in terms of accuracy and interpretability. This work introduces a single-level hyper-box classification approach, which is formulated mathematically as Mixed Integer Linear Programming model. Its objective is to identify the patterns of the dataset using a hyper-box representation. Hyper-boxes enclose as many samples of the corresponding class as possible. At the same time, they are not allowed to overlap with hyper-boxes of different class. The interpretability of the approach stems from the fact that IF-THEN rules can easily be generated. Towards the evaluation of the performance of the proposed method, its prediction accuracy is compared to other state-of-the-art interpretable approaches in a number of real-world datasets. The results provide evidence that the algorithm can compare favourably against well-known counterparts.