Study On Index System And Model Selection In The Unemployment Warning System

At present, China’s GDP growth rate down, the economy into the new normal stage. In the face of the current economic difficulties, the Chinese government put forward to strengthen the reform of the supply side. The supply side reforms to eliminate backward production capacity, will inevitably lead to some employees laid off, and therefore lead to the society on the supply side reforms will induce the "unemployment" intense discussions, American Nobel the economics prize winners Joseph Stiglitz pointed out:without sufficient demand and supply side reforms but will increase unemployment. Whether the supply side reform will lead to a large number of unemployed, the answer is still unknown. But regardless of whether the increased unemployment, build up the early-warning mechanism of unemployment, grasp the dynamics of unemployment, and combining with the data mining technology and mathematical methods, effects of unemployment on the important factors are analyzed, and the unemployment crisis before the outbreak of the beforehand control, undoubtedly has a very important significance.Construction of unemployment early warning indicator system and choose unemployment early warning model is to study the key issues of unemployment early warning stage.The construction of the early-warning index system of unemployment is not perfect in the previous studies, the neglect of the unemployment situation itself and the social security will affect the change of unemployment. This article is selected from the national economic development, labor, and the price of living, social security and unemployment and other aspects of the 24 factors as the primary index, by comparison analysis and correlation analysis of random forests screening index, use random forests screened 10a high degree of correlation index as an argument variable. In the choice of measure index of unemployment, the vast majority of research is the registered unemployment rate in cities and towns, but registered urban unemployment cannot effectively replace the unemployed in society, so this article choose the registered unemployment rate in cities and towns as a measure of unemployment is not reasonable. In this article, we use the implicit unemployment rate formula to calculate the number of hidden unemployment in rural areas, and the government announced the addition of the number of urban unemployment as the number of unemployed. Unemployment this was used as a predictive model of the dependent variable.In the choice of the unemployment early-warning model, the machine learning model is seldom used in the past, and few studies on the effect of each model predictions were compared. In this article, firstly, we construct a multiple linear regression model based on the index data of 1994-2013, and the linear relationship is not significant, then using support vector machine regression, random forest regression and neural network regression three kinds of models that do not consider the relationship between the independent variables of the model to predict the unemployment rate. The results show that the prediction effect of support vector machine regression algorithm is the best, followed by the random forest regression, the prediction effect of the neural network is the worst, and the importance of random forest is also given. Finally, this article uses the diffusion index to judge the change of the unemployment rate, and according to the different warning signals, analyzes the unemployment alarm, provides the theoretical basis for the government decision-making.This article also provides a structured approach to systems development to build the idea of unemployment early warning system. It is divided into information collection, information preprocessing, unemployment forecast, unemployment alerts and expert advice five functional modules. The unemployment early-warning index system and the unemployment forecasting model are applied to the unemployment forecast module, and the unemployment warning module based on the diffusion index is applied in the unemployment warning module.