Abstract: The kernel function of Support Vector Machine（SVM） used of inner product causes “black box” problem. Currently, the research of “black box” mainly adopts the rule extraction method to solve classification problems, but the problem of the regression is rarely mentioned. This paper tries to propose an SVM regression rule extraction method based on the regression tree algorithm. The algorithm takes full advantage of the special properties of support vectors and the advantage of the regression tree, establishes tree model based on support vectors. It successfully extracts high decision-making capacity, including less variables, a small amount of calculation and easy to read rules. Through training and testing the standard data set “Auto MPG” and the actual production data of coal to methanol, comparison of results with other algorithms shows that, the extracted regression rules in the training accuracy and prediction accuracy have a certain degree of improvement.

Key words: support vector machine, rules extraction, regression tree

摘要： 支持向量机（SVM）因为核函数应用内积运算造成了模型较强的“黑箱性”。目前SVM的“黑箱性”研究主要采用规则提取方法解决分类问题，而回归问题鲜有提及。针对回归问题，尝试性提出基于回归树算法的SVM回归规则提取方法，算法充分利用支持向量的特殊性以及回归树的优势，建立支持向量的决策树模型，成功提取出决策能力高，包含变量少，计算量小且容易读取的规则。通过标准数据集Auto MPG和实际的煤制甲醇生产数据集进行了验证，与其他算法对比分析结果表明，所提取的回归规则在训练精度和预测精度等方面都有一定程度的提高。

关键词: 支持向量机, 规则提取, 回归树