Research Of Population Unbalance Risk Early-Warning Model Based On Support Vector Machine

In2011, China’s Twelfth Five Year Plan (2011-2015) put forward the strategic target of "promote the sustainable and balanced development of the people", which makes the population unbalance risk early warning become a new subject in areas of population nowadays.The population unbalance risk has many complex forms. It can be theoretically divided into240kinds and has the characteristic of endogeny, diffusion, and so on. The complexity of the problem enlarges the difficulty in the population unbalance risk early warning. It is hard to reach a satisfactory solution through the traditional methods such as statistics evaluation or system evaluation.This paper researches on population unbalance risk early-warning model based on support vector machine (SVM).The technique of SVM is suitable for small sample, nonlinear and high-dimensional data processing. Furthermore, it has a good generalization performance. All these make it meet the requirements of the population unbalance risk early warning exactly. Firstly, indexes which can characterize the population unbalance risk are abstracted on the basis of analyzing the population unbalance risk points to constitute the population unbalance risk early warning indexes system, their rational threshold values being set. Secondly, with some intelligent computing methods such as SVM, rough sets (RS) and decision trees (DT) being introduced into the population risk early-warning area, a modularized population unbalance risk early warning model is built based on support vector machine technology. One module of this model is a group of standard SVM and RS-SVM, which is used to forecast the warning level of the total risk. The other module is a group of DT-SVM, which is used to forecast the every warning level of the subentry risks. Thirdly, targeting samples difficult to classify, a new verifying model is built based on rough sets technique which uses the warning levels of all subentry risks to forecast the warning level of the total risk. That is, a decision table is set up with the warning level of the subentry risk used as the condition attribute and the warning level of the total risk used as the decision attribute, and decision rules are generated by rough sets algorithm. Then through these rules, samples difficult to classify can be aptly classified, and this result can be used to verify the accuracy of the former result carried out by the RS-SVM. This verifying method also can be used in other areas involving subentry knowledge.