Feature selection aims at selecting important features that can enhance learning performance in data mining, pattern recognition, and machine learning. Filter feature selection methods offer computational efficiency and feature evaluation criteria, while feature interaction information, which may greatly help increase classification accuracy, is often ignored. In this work, we instead propose a novel feature selection algorithm that uses the “maximum of the maximum” criterion to select highly relevant features and their maximally interactive features. Extensive experiments are performed to evaluate the performance of the proposed method with regard to the number of selected features and classification accuracy on thirty UCI datasets. The results demonstrate that the proposed algorithm not only efficiently selects the relevant features and the interactive features, but also enables classifiers to achieve classification accuracy that is better than, or comparably well to, ten representative competing feature selection algorithms.

特征选择旨在选择能够提高数据挖掘、模式识别和机器学习中学习性能的重要特征。过滤器特征选择方法提供了计算效率和特征评估标准，而可能大大有助于提高分类精度的特征交互信息往往被忽略。在这项工作中，我们提出了一种新颖的特征选择算法，该算法使用“最大值中的最大值”标准来选择高度相关的特征及其最大交互特征。进行了广泛的实验以评估所提出的方法在三十个 UCI 数据集上的所选特征数量和分类准确性方面的性能。