SGER: Toward a Unifying Taxonomy for Feature Selection

这个探索性研究（SGER）项目的目标是建立一个统一的特征选择分类法。特征选择用于各种应用中，包括模式识别、机器学习、数据挖掘或决策制定，以在可用的特征中选择最合适的特征子集用于任务。它可以被看作是一个优化问题的指数时间复杂度沿着几个维度。许多特征选择算法已经被开发并部署在现实世界的应用中。然而，存在着什么样的理论建议和什么样的实践揭示之间的一个明显的差距，和扩散的特征选择算法，使它非常难以充分理解的各种特征选择技术，并建立一个通用的方法来进行特征选择。解决这些问题并制定统一的分类法，以促进特征选择方面的新研究、开发和工具，时间紧迫。本项目探讨了处理这些问题的第一步。为特征选择建立统一分类法的任务分为两个步骤：（1）定义一个公共平台，以平等地考虑代表性算法;（2）建立一个统一的分类法，以发现算法如何相互补充以及缺少什么。该方法包括收集有代表性的数据和算法，并进行比较实验，以确定特征选择算法的特性，它们在不同数据和任务上的性能。这个项目的预期结果包括一个当代的调查，一个统一的分类特征选择算法，和一些潜在的解决方案的自动选择问题-能够自动选择最合适的特征选择算法与给定的问题条件。该项目的进展和最新情况以及由此产生的调查和统一分类法将在网上公布。

The objective of this Small Grant for Exploratory Research (SGER) project is to establish a unifying taxonomy of features selection. Feature selection is used in used in various applications, including pattern recognition, machine learning, datamining or decision making, to choose the most appropriate subset of features among the available ones for the task. It can be viewed as an optimization problem of exponential time complexity along several dimensions. Many feature selection algorithms have been developed and deployed in real-world applications. However, there exists a distinct gap between what theory suggests and what practice reveals, and the proliferation of feature selection algorithms makes it very difficult to fully understand the various feature selection techniques and construct a general methodology for feature selection. It is time-critical that these issues are addressed and a unifying taxonomy is developed, to facilitate new research, development and tools in feature selection. This project explores the first step toward dealing with these issues. The task of establishing a unifying taxonomy for feature selection is accomplished in two steps: (1) defining a common platform to consider representative algorithms on the equal footing; and (2) building a unifying taxonomy to discover how the algorithms complement each other and what is missing. The approach includes collection of representative data and algorithms and conducting comparative experiments to determine the characteristics of the feature selection algorithms, their performance on different data and tasks. The expected results of this project include a contemporary survey, a unifying taxonomy of feature selection algorithms, and some potential solutions to the automatic selection problem -- being able to automatically choose the most suitable feature selection algorithm with given the problem conditions. The progress and updates of the project, and the resulting survey and unifying taxonomy will be available online.