Abstract: To solve the efficient and accurate prediction issues for big data,a parameter adaptive Support Vector Regression (SVR) was proposed based on the Mean Shift Clustering (MSC) and Marine Predators Algorithm (MPA).The large sample was divided into training data,validation data and test data.The MSC algorithm was used to deal with the training data for the Cluster Centers (CCs).The SVR parameters were set as kernel function type,penalty factor and kernel function coefficient,and multiple SVR parameter sets were generated randomly.Based on SVR parameter sets and CCs,the SVR was applied to predict the validation data for obtaining the prediction accuracy.The MPA was used to update the SVR parameter sets,and this step was repeated to obtain the optimal SVR parameter set until the cut-off condition was met.The prediction accuracy of the test data was achieved with the SVR and the optimal parameter set.Compared with the highly similar algorithm in prediction accuracy,stability and data loss,the feasibility and effectiveness of the approach were verified.

Key words: big data prediction issue, parameter self-adaption, support vector regression, mean shift clustering, marine predators algorithm

摘要： 为解决大数据条件下的高效精准预测问题,提出一种基于均值漂移聚类与海洋捕食者算法的参数自适应支持向量回归方法。将大数据样本划分为训练组、验证组和测试组;采用均值漂移聚类算法处理训练组得到聚类中心;设定支持向量回归(SVR)参数,随机生成多个SVR参数组;基于参数组和聚类中心,采用支持向量回归算法对验证组样本进行预测以得到预测精度,然后采用海洋捕食者算法更新SVR参数组,循环本步骤直到满足截止条件,从而获得最优SVR参数组;基于该最优参数组,用SVR获得测试组的预测精度。与高度类似方法在预测精度、稳定性和数据损失等方面进行比较,验证了该方法的可行性和有效性。

关键词: 大数据预测问题, 参数自适应, 支持向量回归, 均值漂移聚类, 海洋捕食者算法