基于SVM的大样本数据回归预测改进算法

针对支持向量机回归预测精度与训练样本尺寸不成正比的问题,结合支持向量机分类与回归算法,提出一种大样本数据分类回归预测改进算法。设计训练样本尺寸寻优算法,根据先验知识对样本数据进行人为分类,训练分类模型,基于支持向量机得到各类别样本的回归预测模型,并对数据进行预测。使用上证指数的数据进行实验,结果表明,支持向量机先分类再回归算法预测得到的均方误差达到12.4,低于人工神经网络预测得到的47.8,更远低于支持向量机直接回归预测得到的436.9,验证了该方法的有效性和可行性。

A Modified Regression Prediction Algorithm of Large Sample Data Based on SVM

A modified prediction method of large size data based on Support Vector Machine（SVM） classification and regression is proposed aiming at the problem that prediction accuracy of SVM regression is not proportional to the size of training sample. The method combines the SVM classification and regression algorithms. The size of the sample data is optimized, and the sample data is classified based on a priori knowledge. According to the classification, the classification model is trained. Then it trains the regression model for training sample of all classes, and makes the prediction with large size data based on SVM classification and regression. With the case of Shanghai Composite Index, the Mean Squared Error（MSE） of values predicted by the new method based on SVM classification and regression is 12.4, lower than 47.8 predicted by Artificial Neural Network（ANN） and much lower than 436.9 predicted by SVM regression. These results verify the effectiveness and feasibility of the method.