基于神经网络集成的失业预警方法

提出采用神经网络集成技术对中国失业预警系统进行建模,以克服当前失业预警系统建模中存在的小样本、高维度、非线性、噪音数据等难题。采用BP神经网络回归模型对失业率进行预测;基于两种集成技术Bagging与AdaBoost对多个神经网络进行集成,以获得比单个预测模型更好的精度与稳定性;最后基于广东省的社会经济调查数据进行了实证分析,实验结果表明:在对失业率的预测上,Bagging集成方法的预测效果优于Adaboost集成方法,也优于单个最好的神经网络模型。

Unemployment Early -Warning Based on Neural Network Ensembles

This paper proposes to apply neural network ensembles to unemployment early -warning systems modeling in order to overcome effectively some difficult problems,such as small samples,high dimensions,nonlinearity,noisy data.BP neural networks are utilized as individual forecasters to evaluate unemployment rates,and two ensemble methods,Bagging and Ada-Boost, are used to combine forecasting results of many individual neural networks to obtain better forecasting effectiveness.Experimental results based on social and economic investigation data from Guangdong province show that Bagging produced better effectiveness on forecasting unemployment rates than both AdaBoost and the best individual neural network.