【摘要】 指出失业预警系统的建模是一个小样本、高维度、非线性、存在噪音数据的复杂的建模问题,重点探讨了基于回归分析技术对失业预警系统进行建模的理论、方法与步骤。讨论了常见的缺失数据处理、数据归一化以及特征降维等数据预处理方法;进一步分析了最小二乘回归、Logistic回归、岭回归、BP神经网络以及支持向量回归五种回归技术;最后基于广东省的社会经济调查数据对五种回归方法进行了实证分析,实验结果表明:在对失业率的预测上,支持向量回归预测效果最好,最小二乘回归、岭回归与BP神经网络次之,Logistic回归预测效果最差。更多还原

【Abstract】 Unemployment early-warning system modeling is a special and complex problem with small sample,high dimension,nonlinearity,and noise data.The theory,method,and procedure on modeling unemployment early-warning systems based on regression analysis are mainly discussed.Firstly,the pre-processing methods are given,including missing data processing,data scaling,and dimension reducing.Five regression techniques are further analyzed: least square regression,Logistic regression,ridge regression,BP neural network,and support vector regression.Experimental results based on social and economical investigation data from Guangdong Province show that,support vector regression outperforms the other four methods,logistic regression is worst,and the forecast accuracies of least square regression,ridge regression,and BP neural network are between those of support vector regression and logistic regression.更多还原