基于决策树特征提取的支持向量机在岩性分类中的应用

Application of Support Vector Machine Based on Decision Tree Feature Extraction in Lithology Classification

由于支持向量机属于黑箱模型,因此在进行模型学习时无法直接对特征进行选择,而决策树模型在递归创建的过程中自身具有一定的特征选择能力。针对岩性分类问题,本文将决策树和支持向量机结合,通过决策树的建立,在考虑特征重要性的前提下,利用树节点的高度对特征进行提取,并将具有更高分类能力的特征送入支持向量机进行岩性分类。结果表明:通过决策树的特征提取,减少了支持向量机模型的输入特征,从而有效控制了模型的复杂度,使得模型更加稳定并具有更高的分类精度,测试集精度能够提升10%以上。

Support vector machine is a kind of black box model,and its feature cannot be selected directly when learning model;while decision tree model has the ability of feature selection during the process of recursive creation.For lithology classification,we combined decision tree with support vector machine.In consideration with the importance of the features,we used the tree height to extract the features after the decision tree establishment,and furthermore,we used the features with higher classification ability to fed into the support vector machine.The results show that the feature extraction of decision tree can reduce the input characteristics,so this,in turn,makes the SVM model more stable and accurate through controlling the complexity of the model effectively.The accuracy of test set of the model can be increased by more than 10%.