| With the rapid development of the economy and the steady improvement of the real estate market,the market for second-hand property transactions has expanded annually,which has led to the increasing problem of reasonably assessing the price of second-hand properties.On the one hand,the wide range of features that houses have and the complex link between house prices and features have made it necessary for the process of assessing house prices to take the impact of multiple types of features into account.On the other hand,there are several challenges with the house price valuation process,including valuation being highly influenced by subjectivity,inefficiency in valuation due to the abundance of real estate data,and low explanation of the causes of house price fluctuations.Therefore,a house price valuation method that can effectively estimate the price of a house and has some interpretation of the price components needs to be explored.In the thesis,an urban real estate feature price model based on a hierarchical probabilistic structure is investigated,which builds on data mining techniques,Hedonic Price Model(HPM),and machine learning algorithms.In order to obtain a housing feature dataset,second-hand housing data,Point of Interest(POI)data,school data and public transport facility data are collected and used to construct an urban housing feature dataset containing basic housing features,location features,and environmental features through data mining techniques.The main findings are as follows:(1)A Bayesian probabilistic model for urban house price assessment based on location submarkets is proposed in order to address the problem that characteristic price models tend to have insufficient predictive accuracy and interpretability in assessment.The model draws on the idea of submarket clustering,where a submarket is regarded as a potential variable and submarket criteria are established based on location characteristics;the submarket criteria and the characteristic price model serve as probabilistic dependencies of a Bayesian network that can determine the range of submarket effects;the probabilities and predicted prices of houses in each submarket can be obtained.By using housing data in Hangzhou,the performance of the model algorithm is tested,and the results of submarket segmentation are analyzed.The results show that the model proposed in the thesis has better accuracy in predicting house prices than the comparison model,and that the model has some interpretability,which is feasible in house price assessment.(2)The individual sub-market model updating approach is proposed to solve the problem of updating the characteristic price model for incremental real estate data.The house price assessment model constructed from the previous batch data is viewed as a whole,while the individual sub-market model is viewed as a node.When new second-hand housing data is added,the incremental data set of each sub-market is divided using the location features of the incremental data,and the model is updated for the sub-market to which the incremental data belongs.The results show that the application of the updating method to the incremental housing data not only improves the model prediction accuracy but also reduces the model updating workload.(3)Differences in price models for different city real estate characteristics are compared and analysed,while an urban house price valuation system based on a Bayesian probabilistic model for location submarkets is presented to investigate the applicability and practicality of house price valuation models.For the applicability problem,a Bayesian probability model based on location-based submarkets is constructed using two city datasets with the same characteristics.The model performance differences and the regional composition of submarkets between cities are compared.The findings indicate that when the model is applied,the performance of the algorithm improves in different cities.By analysing the sub-markets,similar key influencing factors are found for the same functional areas in different cities.For the practicality problem,a house price evaluation system is built based on the model in the thesis.The system is built using the Django framework,and the housing data and prediction model are invoked according to user requests,displayed in a visual interface with geographical icons,pie charts,bar charts,etc.,to achieve a modular function with house price evaluation and information query. |
