基于深度学习与随机森林的高维数据特征选择

Feature selection for high dimensional data based on deep learning and random forest

针对特征选择算法对高维大数据降维效果与稳定性差的缺点,提出一种基于深度学习与随机森林的大数据特征选择算法。设计基于随机森林的特征消除算法,对高维大数据集进行特征降维处理;采用保留的特征对受限玻尔兹曼机进行训练,确定受限玻尔兹曼机的模型结构与权重;使用训练受限玻尔兹曼机的学习参数初始化一个多层神经网络,通过标准的后向传播方法训练多层神经网络。基于多组数据集的实验结果表明,该算法提高了高维数据集特征选择的化简效果,保持了较高的稳定性与鲁棒性。

Aiming at the disadvantages of poor effects of dimensionality reduction and stability of feature selection algorithms for high dimensional big data,a feature selection algorithm for high dimensional data based on deep learning and random forest was proposed.A feature elimination algorithm based on the random forest was designed to realize the feature dimensionality reduction of high dimensional big data set.The preserved features were adopted to train the restricted Boltzmann machine,and the model structure and weights of restricted Boltzmann machine were decided.Learning parameters of the restricted Boltzmann machine training were used to initialize a multi-layer neural network,and the multi-layer neural network was trained through the standard back propagation method.Experimental results based on several data sets show that the proposed algorithm improves the effects of dimensionality reduction,and it also shows good stability and robustness.