Research On The Development Level And Dynamic Clustering Of National High-tech Zones

Since the establishment of China’s first high-tech zone in 1988,China’s national high-tech zone has developed rapidly in the past 30 years.All provinces have established their own high-tech zones.However,due to the vast territory of China,there is a great gap between the economic development levels of various regions or provinces.The gap is extremely uneven.In the past,research on national high-tech zones was mostly based on data from a certain year,and cluster analysis was usually used to study the development level and status of each high-tech zone.Although the previous research methods have achieved good application results and can provide some effective policy recommendations for national policies,all previous methods have neglected an important fact that there is a great relationship between the economic development levels of various regions or provinces in China.The big differences,which in turn lead to the original investment in the construction of high-tech zones,including human,financial and policy openness,will also vary greatly.Without abandoning the original input,only the total analysis method can not make a reasonable prediction of the real development of the high-tech zone,and thus will give unreasonable policy recommendations.In view of this,this paper proposes cluster analysis based on the raw data and factor analysis data of relative development rate indicators.Further,this paper will study the automatic selection problem of the number of clusters,which is different from supervised learning such as regression or classification.Cluster analysis belongs to unsupervised learning,so that the number of clusters cannot be selected by cross-validation commonly used in regression or classification.This paper innovatively introduces 26 methods for selecting the number of clusters in previous studies,such as CH(Calinski and Harabasz)methods,Duda method,etc.,using the similar classification voting principle to select the most selected clusters among the 26 methods as the final cluster.number.Finally,the clustering results based on the relative development rate indicators are compared with the data clustering results based on 2014 and 2017,respectively.The clustering results based on the data of 2014 or 2017 are similar,but the two are similar.However,theclustering results in the past two years are quite different from the results based on the relative development rate indicators,namely the dynamic clustering proposed in this paper.Combined with the annual development status,high-tech zones with real high-speed development can be found.Therefore,this paper suggests that the government should further increase investment in high-tech zones that are truly developing at a high speed,and reduce the investment in high-tech zones that are only due to abundant raw inputs but slow in development,so as to achieve optimal allocation of resources.