基于单信息初始搜索的特征选择方法研究

Research of Feature Selection Method Based on Single Information Initial Search

在机器学习任务中,特征选择是重要的数据预处理,可为获得较好的特征数据集,有利于训练产生精确度、可靠性等适应能力较强的学习模型.通过不同的评估策略,应用多种特征选择方法挖掘出有利学习模型的特定数据集,提出了基于单信息特征评估策略作为搜索特征子集的初始方法,并结合典型特征选择方法进行比较研究,实验结果表明该方法可提高分类的运行效率和准确度.

In the machine learning task feature selection is an important data preprocessing, and a better feature data set can be obtained, which is beneficial for training to generate a learning model with strong adaptability and accuracy. Through different evaluation strategies, a variety of feature selection methods are used to mine specific data sets of favorable learning models. An initial method based on single feature evaluation strategy as a subset of search features is proposed and compared with typical feature selection methods. It shows that this method can improve the classification efficiency and accuracy.