基于随机森林优化的自组织神经网络算法

针对基于降维的神经网络分类器预测模型在分析过程中存在特征丢失, 并导致精度下降的问题， 提出一种基于随机森林算法优化的多层感知器(MLP)回归预测模型. 该优化模型通过在MLP回归模型网络的全连接层和逻辑回归层之间增加一个优化机制, 利用随机森林算法对隐藏层状态的优化实现改进, 从而解决了降维过程中神经网络丢失数据特征的问题. 在借贷客户信息数据集上的实验结果表明, 该模型在保证主要特征的同时大幅度提升了预测准确率, 证实该模型在特征工程中具有较高的实用性.

Self-organizing Neural Network Algorithm Based on Random Forest Optimization

Aiming at the problem of the loss of features and precision degradation in the analysis process of neural network classifier prediction model based on dimension reduction, we proposed a multi-layer perceptron (MLP) regression prediction model optimized by random forest algorithm. The optimization model added an optimization mechanism between the full connection layer and the logistic regression layer of MLP regression network, the random forest algorithm was used to optimize the state of hidden layer, so as to solve the problem of losing some data features in the process of dimension reduction of neural network. The experimental results on the information data set of the borrowing customers show that the model can guarantee the main features and greatly improve the prediction accuracy, which proves that the model has high practicability in feature engineering.