基于特征选择的极限随机森林算法研究

高维复杂数据处理是数据挖掘领域中的关键问题，针对现有特征选择分类算法存在的预测精确度失衡、整体分类效率低下等问题，提出了一种结合概率相关性和极限随机森林的特征选择分类算法（P-ERF）。该算法使用充分考虑特征之间相关性与P值结合的特征选择方式，避免了树节点分裂过程中造成的冗余性问题；并以随机树为基分类器、极限随机森林为整体框架，使P-ERF算法获得了更高的精准度和更好的泛化误差。实验结果表明，P-ERF算法相较于随机森林算法、极限随机森林算法，在数据集分类精度与整体性方面均得到良好的效果。

Research on extreme random forest algorithm based on feature selection

High-dimensional complex data processing is an important problem in the field of data mining. Aiming at the problems of imbalance of prediction accuracy and low overall classification efficiency of existing feature selection integrated classification algorithms, this paper proposed a feature selection classification algorithm combining probability correlation and extreme random forest. The algorithm used a feature selection method that fully considered the correlation between the features and the P-value, which avoided the redundancy caused by the tree node splitting process; Besides, this algorithm took the random tree as the base classifier and the extreme random forest as the whole framework, which made it achieve higher accuracy and better generalization error. The experimental results show that the algorithm can achieve good results in data set classification accuracy and integrity compared with random forest algorithm and extreme random forest algorithm.