Exploration And Realization Of Land Price Classification Model Based On MapReduce

The price of land is more and more significant with the development of urbanization process and the increase of marketization degree of land.Not only do financial practitioners or employees of an estate enterprise pay attention to the land price,but almost everyone cares about it.A reasonable land price can control the scale of a city and optimize the industrial structure.It can also promote the process of urbanization and enhance city's competitiveness.With the development of photogrammetry technology,computer science and network,the data volume is increasing.Facing a large number of spatial data,normal PCs encounter the bottleneck when processing data,and we need an efficient and stable scheme to put the treating process into cloud.We can use Hadoop to solve the problems about calculation,data collection and data mining,and this is the innovation point of my thesis.This thesis is about land price classifying and map algebra using MapReduce.It contains such 3 primary aspects.1.Designing distributed algorithm about geospatial raster data using Map Reduce.Referencing technical documentation of ArcGIS and some other open source GIS softwares,I designed some distributed algorithms to deal with problems about geospatial raster data like gradient,Euclidean distance and kernel density.2.Data acquisition and data preprocessing.The range of data covers main urban area of Chengdu,and the data contains land transaction data,interest points,DEM,road data,river data,bus stations,metro stations and so on.The data is mainly obtained through two channels,one is the existing data of the laboratory,the other is gathered by a crawler.When writing crawler,distributed asynchronous crawler was developed in combination with MapReduce,thread pool and other technologies,which improved the efficiency of data crawler.When preprocessing influence factors data,the distributed spatial data computing tool developed in the first step of research was used to generate the land price data set for classification.3.Using the classification model to classify land price data sets.By learning mature classification algorithms like Bayesian classification,decision-making tree classification,random forest,neural network,support rector machine,an appropriate method is chosen to study the implementation strategy under the distributed condition,and the comparison is made under the condition of single machine and distributed condition.Then the land price data set was put into the model for training and back measurement,and the results were analyzed.Finally,the effects on classification accuracy and training efficiency were explored from the perspectives of macro factors supplementing data set and correlation analysis simplifying data set.